For Rebel Wolves, the Warsaw studio built by Witcher 3 director Konrad Tomaszkiewicz and a core team from CD Projekt Red, September 3 is the day four years of development face their first commercial test.

The Studio Born from “Crazy Ideas”

Tomaszkiewicz spent 17 years at CD Projekt Red, directing The Witcher 3: Wild Hunt and heading production on Cyberpunk 2077, before leaving in May 2021 after an internal investigation found no evidence supporting conduct allegations against him. His brother Mateusz, quest director on much of the Witcher 3 campaign, followed. Narrative director Jakub Szamalek, who co-wrote the principal Cyberpunk storyline and much of the Witcher DLC, signed on alongside design director Daniel Sadowski and art director Bartłomiej Gaweł, each carrying years of credits across both franchises.

His reason for starting over was blunt. “We had crazy ideas,” he told PC Gamer at the game’s January 2025 reveal. “We knew that if we wanted to realise them, we had to start our own studio, because it would be hard to convince any big company to change their approach, to do something new and crazy.”

Rebel Wolves announced in February 2022 with roughly two dozen members drawn from CDPR. The team grew to around 90 by May 2024, then reached 160 employees by April 2026 as the game entered certification on PS5, Xbox, and PC. NetEase provided the backing, described at the time as “the last missing piece needed to go full throttle.” Holding no licensed IP puts the full creative and commercial ownership of the Dawnwalker Saga in Rebel Wolves’ hands alone.

A 21-minute pre-beta gameplay overview released after the game’s 2025 Xbox Games Showcase appearance crossed a million views across platforms within days. Community reaction focused on two things: enthusiasm for the atmospheric Carpathian dark fantasy setting and direct comparisons to the Witcher franchise, alongside calls to tighten the camera distance and sword combat feel before release. Rebel Wolves confirmed the September date at their Road to Launch event in late April, where the studio also shared a story trailer and a cinematic teasing the saga’s future direction.

What Pre-Orders Get You

The Blood of Dawnwalker launches in four editions. All pre-orders include the Sangoran Wayfarer’s Armor Set as an early unlock; Rebel Wolves confirmed the armor can be obtained through normal gameplay as well, so nothing is permanently gated behind a pre-order payment.

UK pricing is £59.99 for Standard and £69.99 for the Eclipse Edition on the Xbox Store. Play Anywhere is confirmed on the game’s official Xbox Store listing, covering both Xbox Series X|S and PC through the Microsoft Store with a single digital purchase.

The Thirty-Day Clock

How Time Moves

Coen has 30 in-game days and nights to defeat the vampire lord Brencis and protect his family. Vale Sangora, the hand-crafted open world, takes in forests, plains, swamps, mountain ridges, and walled medieval settlements built around the faction conflicts driving the story. The 14th-century backdrop carries active chaos: Black Death sweeping the region, vampires using the weakened human world to seize power from the shadows. Moving through all of it costs nothing on the calendar.

Time advances only through deliberate choices: completing quests, taking certain dialogue branches, or purchasing upper-tier skills. Every time-consuming activity flags its cost before the player commits. “It will always be clearly telegraphed to the player if an activity moves time forward (and by how much),” confirmed Mateusz Greiner, senior PR manager at Rebel Wolves. Lead quest designer Rafał Jankowski described the structural consequence: “In a way it made many of the game’s stories optional. So we were able to include many different endings and solutions to those stories, all of which can still allow you to finish the game at a certain point.” One playthrough can’t complete every questline.

The full story runs 30 to 40 hours, per developer interviews reported by GamingBolt. With 30 calendar days available, the budget is calibrated to feel meaningful without being punishing.

What Happens When It Runs Out

Running out of days before defeating Brencis doesn’t end the game. The world changes instead. Rebel Wolves stated the outcome plainly: “Coen’s family might suffer the consequences.” Family members can die permanently within that playthrough. Side-quest NPCs who weren’t helped before their questlines expired vanish from the world, including merchants and service providers tied to them.

We don’t want to punish you for doing stuff. So we aimed to get it to the perfect point where you’re not stressed about the time, but still feel that this big event is coming closer and closer.

The studio’s CEO and game director said that during a PlayStation Blog preview session ahead of the release date announcement. Design director Daniel Sadowski described the system’s purpose: it “will definitely force you to make choices at some points, like what to do and what to ignore to maximize your chances of defeating the main enemy.” A player can head straight for Brencis without most side content; the final encounter scales to how prepared Coen arrives.

Sword at Dawn, Fangs at Dusk

The day and night divide cuts Coen’s toolkit completely in two, with each mode carrying combat abilities, traversal options, and accessible world areas the other doesn’t reach.

During daylight, Coen is an infected human. Combat relies on swordfighting backed by Hex magic, with a directional blocking system requiring players to match one of four parry angles. Damage builds activation charges that trigger execution strikes or release Hex magic in concentrated bursts. Blood-magic sites across Vale Sangora are accessible only in daylight and lock completely at sunset.

After dark, the vampiric vrakhiri skill tree takes over. Wolf-form transformation opens for rapid traversal, and supernatural strikes unavailable in daylight become active. An infamy system tracks aggressive public actions and adjusts how factions within the vampiric court respond to Coen across the map; spiked infamy can close allied NPC options that don’t reopen. A blood hunger mechanic accompanies the nighttime form as well, pressuring Coen to feed during exploration and adding friction to extended night operations.

Three branches organize the full skill progression:

• Witchcraft: daytime Hex magic, activation charges, and access to blood-magic sites
• Vampiric powers: the nocturnal vrakhiri tree, wolf-form traversal, and supernatural combat strikes
• Swordfighting: active across both modes, the base combat layer for day and night play

For motion capture on Bakir, one of Brencis’s elite vampire lieutenants, Rebel Wolves brought in Jan Błachowicz, former UFC Light Heavyweight Champion (title held 2020 to 2021). The goal was combat that carries the weight of professional martial arts movement rather than floating choreography.

The Witcher Parallel Rebel Wolves Invites

Comparisons to the Witcher franchise were immediate after the first gameplay footage surfaced. Tomaszkiewicz directed The Witcher 3. His brother Mateusz was its quest director. Szamalek wrote large portions of the DLC. The thematic ground, dark fantasy medieval Europe with player-driven moral weight, runs close. The studio has never tried to push the comparison away.

“I’m really comfortable. I feel really confident about our game; we’re already past friends-and-family testing and focus tests at professional companies, and the feedback is really good,” Tomaszkiewicz said in a pre-release interview ahead of the launch event. CD Projekt Red developers were publicly praising Dawnwalker during the same period, and Rebel Wolves returned the sentiment toward The Witcher 4, choosing co-existence over rivalry.

The mechanical structure is a Persona model applied to a dark fantasy open world. Atlus’s modern Persona games built replayability on calendar commitment: time moves, choices are permanent, no single run completes the picture. Rebel Wolves has confirmed 100 percent completion in one run is impossible by design, with the replayability structure coming from the time budget itself rather than from post-credits content or difficulty modes.

IP ownership separates the studios further. The next Witcher and Cyberpunk franchise entries carry constraints accumulated over decades of licensed history. Rebel Wolves holds the full Dawnwalker Saga. At the Road to Launch event, a cinematic teased a future installment with Coen in a modern-day setting; the studio’s co-founder was careful to frame each entry as standalone. “Each installment in the series will be a standalone game with its own story,” he said, “without narrative cliffhangers, unresolved plot threads, or time travel ideas.”

Pre-orders across all four editions are live on the Xbox Store now, and the game ships on September 3.

The case rests on more than a single quarter’s numbers. The five largest cloud providers finished Q1 with a combined $2.1 trillion in signed contracts for future services, up 184% year over year. Between $400 billion and $450 billion in new commitments arrived each of the last three quarters, per company filings cited in the note.

Four Consecutive Quarters of Record Sequential Gains

Amazon, Microsoft, and Google generated a combined sequential revenue addition of $6.5 billion in Q1 2026, up 196% from the same period a year earlier. All three have now posted four straight quarters at what UBS called historically elevated sequential gains. The firm expects Q2 2026 to add $8.5 billion in combined sequential revenue, with Wall Street’s own consensus for the remaining quarters running between $8.1 billion and $9.8 billion per quarter. The three clouds broke differently on year-over-year growth rate.

• Google Cloud: 63% year-over-year growth, reaching $20.03 billion in quarterly revenue, a 35-percentage-point acceleration from Q1 2025 and the first time the division has cleared $20 billion in a single quarter.
• AWS: 28% year-over-year growth, reaching $37.59 billion, an 11-percentage-point acceleration over the same 12-month stretch and its fastest quarterly growth rate in 15 quarters.
• Azure: approximately 39% year-over-year growth, held near that pace for four consecutive quarters as Microsoft redirected available capacity toward its own AI products, research, and server migration workloads.

Google’s jump from 48% in Q4 2025 to 63% in a single quarter is the largest sequential acceleration in the group this cycle. AWS moved 11 percentage points over four quarters. Azure’s flat line traces a supply decision. Amy Hood, Microsoft’s chief financial officer, said on the Q3 fiscal 2026 earnings call that customer demand continued to exceed supply, and that Microsoft was balancing new infrastructure deliveries against first-party applications, research, and end-of-life server replacement. She projected modest acceleration in Azure’s growth in the second half of calendar 2026 as capacity deliveries increased.

A $2.1 Trillion Committed Backlog

The aggregate figure covers Amazon, Microsoft, Google, Oracle, and Meta, per company filings cited by UBS. The three largest individual positions break down as follows.

Amazon’s Trainium (Amazon’s custom chip built for large-scale AI training and inference) concentration is the most specific figure in the table. Chip commitments account for nearly two-thirds of the company’s entire cloud backlog, a shift that reflects how quickly enterprise AI procurement moved from general-purpose compute to purpose-built AI hardware over the past 18 months. Amazon CEO Andy Jassy separately confirmed on the Q1 earnings call that the $364 billion figure excluded a new $100 billion-plus commitment from Anthropic made during the same quarter.

Microsoft’s commitment was up $2 billion from the prior quarter, meaning new business continued to accumulate even as the existing stack compounded. Google’s position surged $222 billion sequentially in Q1, its largest single-quarter jump, driven partly by TPU (Tensor Processing Unit, Google’s proprietary AI chip) hardware sale agreements with external customers. UBS characterized the $400 billion to $450 billion in new quarterly commitments across the group as evidence that demand is still accelerating into the infrastructure build.

Azure Tops the Enterprise Preference Survey at 69%

Of the 140 enterprises UBS surveyed, 69% named Azure as their preferred platform for AI infrastructure, with AWS following at 53% and Google Cloud placing third at 27%, down from 31% in an October 2025 survey round. Because enterprises run workloads on multiple clouds simultaneously, these figures don’t add to 100%, but the Azure margin over the field has held consistent across each round UBS has published.

The $37 Billion AI Revenue Run Rate

Microsoft’s AI business crossed an annual revenue run rate of $37 billion in the quarter, up 123% year over year, according to Microsoft’s Q3 fiscal 2026 earnings filing with the SEC. The figure covers AI services running on Azure, revenue from model builders on the platform, and Microsoft’s own tools including Copilot, and excludes workloads drawing only on standard compute or storage. Microsoft 365 Copilot commercial seats crossed 20 million in the quarter, up 250% year over year. Microsoft had disclosed a $13 billion AI run rate in January 2025, making the 18-month move to $37 billion the only public milestone the company has updated on that trajectory.

The $37 billion also excludes revenue from Microsoft’s own MAI model family, the seven in-house AI models unveiled at Build 2026 following Microsoft’s renegotiated OpenAI exclusivity deal, now available to Azure customers through the Foundry platform. AWS reported AI revenues exceeding $15 billion annualized, up more than 100% year over year. Google cited roughly 800% growth in enterprise AI revenues for the quarter, driven by Gemini Enterprise and generative AI model adoption, without disclosing an absolute figure.

Azure’s Supply Constraint and the Preference Lock-In

Hood’s Q3 call cited first-party applications, research, and end-of-life server replacement as competing priorities alongside external cloud demand, a balance she said would hold at least through the end of calendar 2026. External Azure customers are, in effect, competing with Microsoft’s own infrastructure priorities for the same physical capacity.

The UBS survey result sits alongside that constraint. Enterprises named Azure as preferred at 69% after rationing had already begun, pointing to future-contract commitments rather than platform switches. Microsoft’s $627 billion grew by $2 billion in the same quarter the quarterly growth rate held flat.

Google Cloud’s Acceleration and the TPU Hardware Play

Google Cloud crossed $20 billion in quarterly revenue in Q1 2026. Alphabet’s Q1 2026 investor disclosure put the cloud backlog at more than $460 billion, with approximately 50% expected to convert to revenue within the next 24 months. The UBS note reports the precise figure as $462 billion, nearly doubling in a single quarter.

A significant portion of that jump came from TPU hardware sale agreements with external customers. Google historically reserved TPU capacity for internal infrastructure; making the hardware available to outside buyers under contract is a new revenue line for the division. Alphabet CFO Anat Ashkenazi told investors that TPU agreements are included in the $462 billion figure, that a small fraction would convert to revenue later in 2026, and that the majority would be recognized in 2027.

Cloud operating margin reached 33% in Q1, up from operating losses as recently as 2022. Google Cloud ran from loss-making to 33% in four years, the fastest margin progression in the three-cloud group over that stretch, while also scaling quarterly revenue past $20 billion for the first time. The division still trails AWS’s 38%, with the gap narrowing across consecutive quarters.

Alphabet CEO Sundar Pichai said the company remains compute-limited, allocating capacity carefully between internal products and external cloud customers. Enterprise AI offerings became the primary growth driver for the division in Q1 for the first time, with generative AI product revenue up roughly 800% year over year, led by Gemini Enterprise.

Margins Steady at 37% Despite $673 Billion in Capex

Cloud Margins at 37%

The combined operating margin for AWS, Microsoft’s Intelligent Cloud segment, and Google Cloud held at 37% in Q1 2026, unchanged from Q4 2025. AWS ran at 38%. Google Cloud reached 33%, up from 18% in Q1 2025. Microsoft’s Intelligent Cloud segment ran slightly below the group composite, with the overall figure holding flat despite higher AI infrastructure depreciation charges weighing on both AWS and Microsoft, per the UBS note.

Together, the three businesses generated roughly $31 billion in combined operating income at that margin on the $84.8 billion in combined cloud revenue. Google Cloud’s progression from operating losses in 2022 to 33% in four years is the fastest in the group, coinciding with the division scaling from sub-$10 billion to over $20 billion in quarterly revenue.

The $673 Billion Capital Expenditure Build

UBS estimates the five providers will spend a combined $673 billion in capital expenditures in 2026, up 76% from $382 billion in 2025. Alphabet raised its 2026 capex guidance to between $180 billion and $190 billion and said 2027 capex would "significantly increase" from that level. On the Q3 fiscal 2026 Microsoft earnings call, Hood said capital expenditures would exceed $40 billion in the current quarter, putting the company on pace for roughly $190 billion for calendar 2026.

Bank of America has estimated the five largest capex spenders will consume about 90% of their combined operating cash flow on capital expenditures in 2026, up from 65% in 2025. Morgan Stanley projected their combined borrowing to exceed $400 billion this year, more than double 2025’s $165 billion. Free cash flow for the group fell to $200 billion in 2025 from $237 billion in 2024.

The Backlog as UBS’s Answer

Cap-ex keeps climbing, but ROI is evident via ~$2 trillion backlog and accelerating cloud growth.

Jefferies analysts wrote that in a note to investors published after the Q1 2026 earnings cycle. BMO Capital Markets separately cited the backlog as the structural support for what it described as the capex super-cycle.

AWS Bedrock (Amazon’s managed AI development platform) reached 125,000 customers and nearly 80% of the Fortune 100 in Q1 2026, with customer spending growing 170% quarter over quarter, per Amazon CEO Andy Jassy’s Q1 2026 earnings commentary on the company’s official newsroom. The platform processed more tokens in Q1 than in all prior years combined.

Alphabet’s Q1 2026 Form 8-K confirmed that just over 50% of Google Cloud’s $462 billion backlog converts to revenue within 24 months, with TPU hardware commitments converting primarily in 2027.

Disclaimer: This article covers investment research published by UBS and is intended for informational purposes only. It does not constitute financial advice or a recommendation to buy or sell any security. Past performance is not indicative of future results. Readers should consult a qualified financial adviser before making investment decisions. All figures were accurate as of publication.

Matt Booty, Xbox’s chief content officer, confirmed before today that the event would focus on titles releasing within the next 12 months, a deliberate turn away from the multi-year pipeline reveals that have padded prior showcases. Project Helix, the next-generation Xbox console, won’t appear. What will appear is the most concrete picture yet of Sharma’s Xbox software slate heading into the back half of 2026 and into 2027.

The Leadership Reset Behind Today’s Event

Spencer’s decade at the top of Xbox included the Game Pass buildout, the $68.7 billion Activision Blizzard acquisition, and a broad push toward platform-agnostic publishing that increasingly put Xbox games on PC, cloud, and competing consoles. The final years also brought significant studio cuts. Arkane Austin, developer of Redfall, closed in May 2024, the same month Tango Gameworks, which had just shipped Hi-Fi Rush to strong critical reception, was also shut down. Those decisions landed after years of Xbox arguing its acquisition strategy would produce a richer game portfolio, and the tension between that promise and those outcomes is part of what Sharma stepped into.

Her stated direction is explicitly brand-focused. The language she has used publicly centers on “a renewed commitment to Xbox,” which is tighter than the platform-agnostic framing that defined Spencer’s final years. Concrete changes came quickly: Game Pass subscription prices were lowered, the Xbox boot-up sound was updated, and the Achievement system was overhauled in the weeks following her appointment.

The first public test of her instincts didn’t land cleanly. Fable’s delay was announced shortly before today’s show, and Booty followed it with an appearance on the Official Xbox Podcast where he confirmed that multiplatform logos, PlayStation 5, Switch 2, Steam, would still appear in the broadcast. Community pushback followed. Sharma addressed it directly from her verified X account on May 29:

Seeing the feedback on logos. It was a miss, and I own it. We are talking about how we adjust for future XBOX shows.

Those logos will be visible today. On the hardware front, Booty confirmed Project Helix absent from the showcase entirely; the next-gen console has no confirmed launch window, and the 12-month software focus is the framing the new CEO wants today’s broadcast to carry.

Start Times and Where to Stream

Both the main showcase and the E-Day Direct stream across channels listed in Xbox Wire’s official broadcast guide for today’s double feature. Global start times for the main show:

• 10:00 AM PT (Pacific Time)
• 1:00 PM ET (Eastern Time)
• 6:00 PM BST (British Summer Time)
• 7:00 PM CEST (Central European Summer Time)
• 3:00 AM JST (Japan Standard Time, June 8)
• 1:00 AM AWST (Australian Western Standard Time, June 8)

US viewers get a new distribution option this year: Amazon Live and Prime Video on Fire TV. The YouTube stream runs in 4K at 60 frames per second. Accessibility formats running simultaneously include English Audio Description on Xbox’s main YouTube channel, American Sign Language on that same channel and on the dedicated /XBOXASL Twitch channel, and British Sign Language on Xbox On’s YouTube.

The E-Day Direct follows the main broadcast immediately. Prior Developer Directs attached to Xbox showcases have averaged around 45 minutes. Last year’s equivalent, a standalone show for The Outer Worlds 2, began within seconds of the main event ending.

Gears of War: E-Day Gets a Dedicated Direct

Among the games on the board today, only one came with a named post-show segment confirmed in advance by the Xbox Wire announcement for today’s double feature: Gears of War: E-Day. The Coalition and People Can Fly are co-developing it, a pairing with franchise history behind it. The Coalition built Gears 4 and Gears 5. People Can Fly co-developed the original Gears trilogy and later led Gears of War: Judgment before departing to build its own projects for several years.

The game runs on Unreal Engine 5. The Coalition’s technical director Kate Rayner said E-Day has over 100 times more environment and character details than Gears 5, which used Unreal Engine 4. Hardware-accelerated ray tracing, including ray-traced reflections and shadows, is confirmed alongside a new destruction and gore system. Brand director Nicole Fawcette described Marcus and Dom in early Gears titles as “big beefy tanks, exaggerated in their design” because of technical limits, and said they have “come to life as human characters” in E-Day through modern rendering. The design direction also returns to a linear structure, pulling back from the open-world elements in Gears 5 that The Coalition has since described as a departure from what the franchise does best.

It’s the sixth mainline Gears title and a prequel set 14 years before the original game. The story follows younger versions of Marcus Fenix (voiced by John DiMaggio) and Dom Santiago in Kalona, a new city in the Gears universe, on Emergence Day: the moment the Locust Horde surfaces from underground and collapses civilization. The game was first announced via in-engine trailer in June 2024 and targets a 2026 release on Xbox Series X/S and Windows.

The Direct will offer, per Xbox Wire, “new details, gameplay and insights about the hugely anticipated origin story to the Gears of War saga.” Two years have passed since the initial reveal, and whether the segment closes with a specific launch date is the clearest open question heading into it.

The Lineup Beyond E-Day

Several other titles carry official confirmation or strong pre-show evidence across Microsoft’s first-party studios and its third-party publishing slate.

Near-Confirmed Appearances

Fable is confirmed to appear. Playground Games’ reboot of the British fantasy RPG was officially pushed to February 2027, with Xbox describing a “major new look” at today’s showcase. The February window has practical logic beyond the stated reasoning: GTA 6 is scheduled for November 19, and clearing that calendar space entirely gives Fable its own launch window.

Halo: Campaign Evolved is a remake of the original Halo: Combat Evolved campaign from Halo Studios (formerly 343 Industries), scheduled for a 2026 release. Details leaked briefly from the Microsoft Store before being pulled, showing three new prequel missions set before Master Chief arrives at Installation 04, Alpha Halo. One screenshot appeared to depict Master Chief piloting a Covenant Seraph fighter above Reach, a type of space-combat sequence the franchise has never offered in a playable FPS context before. Pre-release purchase listings with armory packs were also visible before the pages came down.

State of Decay 3, from Undead Labs, was announced nearly six years ago at the July 2020 Xbox Games Showcase and has had minimal public visibility since. In April, Undead Labs opened alpha test registrations, and a limited play session ran in May. Windows Central’s Jez Corden reported the game is already in private alpha testing and said it sounds like the studio “really nailed the ‘AAA’ feel that was absent from State of Decay 2.” Wccftech rated its appearance today at 80 percent confidence. Clockwork Revolution, from inXile Entertainment, is the first-person action RPG set in an alternate-history steampunk city with time-manipulation mechanics. It has appeared at two consecutive prior Xbox summer showcases without a release window; a third showing would almost certainly bring one.

Metro 2039, announced by 4A Games in April with a late 2026 release slot, has a launch window close enough to today that extended gameplay is a reasonable showcase inclusion. Call of Duty: Modern Warfare 4 is confirmed for at minimum a DMZ extraction mode first look; Activision’s own reveal trailer for the game, published May 28, explicitly advertised “a DMZ first look on Xbox Games Showcase” for June 7.

Rumors With Evidence

Persona 6 is the most-discussed unconfirmed game ahead of the broadcast. Leaker Nate the Hate, with a strong track record on Atlus announcements, predicted the reveal. In the days before today’s showcase, SEGA began issuing copyright strikes against accounts posting Persona 6 leaks, a pattern that typically accompanies active IP protection ahead of an imminent official announcement.

Minecraft Dungeons 2 surfaced on the Nintendo eShop ahead of any official reveal, exposing its release date and pricing early. Bethesda could contribute a Fallout 3 remaster; the concept has circulated in rumors since the Oblivion Remastered shadow drop earlier this year established a usable template. Arkane Lyon’s Blade game, confirmed in full production since 2024, is a likely candidate for footage after an extended public absence.

Call of Duty Breaks With Its Game Pass Pattern

Infinity Ward announced Modern Warfare 4 on May 28 with a confirmed launch on Xbox Series X/S, PlayStation 5, PC via Battle.net, Xbox on PC, and Steam, and on Nintendo Switch 2. Infinity Ward is leading development alongside ten additional studios including Treyarch, Sledgehammer Games, and Raven Software. The campaign runs two parallel arcs: Private Park, a young South Korean conscript fighting a North Korean invasion, and Captain Price, operating outside official military sanction after killing a US general in the events of Modern Warfare III. The story spans Korea, New York, Paris, and Mumbai. Multiplayer ships with 12 six-on-six maps at launch and the returning DMZ extraction mode, in which operators work behind enemy lines recovering technology from contested zones.

Both the Xbox Wire campaign announcement for Modern Warfare 4 and the official Call of Duty blog confirmed the October 23, 2026 launch date. Standard Edition pricing is $69.99; the Vault Edition runs $99.99. Pre-orders opened May 28 across Xbox, PlayStation, Battle.net, and Steam.

What sets this year’s Call of Duty apart from recent entries: unlike Black Ops 6 and Black Ops 7, which both launched into Xbox Game Pass on release day, Modern Warfare 4 will arrive on the subscription service at least a year after its October release. Microsoft announced that shift in April, the same period it reduced Game Pass subscription prices. Anyone holding a Game Pass subscription expecting to play MW4 at launch will need to purchase it separately.

The commercial context matters. Black Ops 7 recorded the franchise’s worst annual sales chart ranking since 2008. With MW4 targeting a premium October window and GTA 6 arriving November 19, Activision has clear reason to protect day-one purchase revenue on what needs to be a recovery entry. Day-one access to the annual Call of Duty was among the most frequently cited subscriber benefits Microsoft used in making its case for the Activision acquisition, cited publicly and in regulatory proceedings across multiple territories. The October 23 release removes that benefit for the first time since the deal closed.

Modern Warfare 4 on October 23 will be the first Call of Duty since that acquisition to require a full purchase on launch day.

The retrieval component, Web IQ, works with any reasoning engine a developer already runs, including models outside Microsoft’s catalog entirely. A team using a self-hosted open-source model can send retrieval queries through Web IQ and receive web-grounded responses without migrating to a Microsoft model first.

The Context Gap Problem

Microsoft’s central argument at Build 2026 was that model quality has stopped being the constraint. Every new agent spins up knowing nothing about the organization it’s supposed to serve. Which employee handles procurement approvals, what the actual quarterly margin figures say, what surfaced in this week’s project review: none of that is available without a bespoke data integration. Run a dozen agents across a large enterprise and you’ve inherited a dozen separate retrieval pipelines, each one a maintenance obligation the moment an underlying system changes.

The Azure blog published alongside the announcements frames the constraint directly:

The challenge is no longer model capability, but consistent, shared data context across the business. Every new agent starts from zero, relearning how the business works, where data lives, and what rules to follow. Without a consistent foundation, agents can’t coordinate or scale.

Microsoft authored that framing, so take it as self-interested. The underlying problem it describes is documented across enterprise deployments: multi-agent systems with no shared data foundation return conflicting answers and duplicate retrieval work. As covered in the Build 2026 preview on WinAddons, this year’s conference catalog was built almost entirely around the agent infrastructure questions — tooling, pricing, governance — that determine whether agents reach production at all. IQ is Microsoft’s consolidation answer: a shared intelligence foundation that agents query from the start rather than rebuild each time.

Four Layers, One Intelligence Stack

Microsoft IQ bundles four data layers, each feeding agents a distinct type of context. They work as a unified stack but can each be accessed independently depending on what a given agent actually needs.

Work IQ’s API surface is deliberately narrow. The official Work IQ API announcement describes the layer as building “a semantic understanding of your business by continuously processing content from email, calendar, meetings, chats, files, people, collaboration patterns, and your line of business systems.” The retrieval surface collapses to ten generic tools exposed via the Model Context Protocol (MCP, the open standard for connecting AI models to external data sources), so agents need not learn hundreds of endpoint-specific calls. Those APIs go generally available on June 16, 2026.

Fabric IQ connects to OneLake and gives agents queryable access to actual analytics figures rather than summaries inferred from stale documents. An agent asking about quarterly revenue hits the real data store. Foundry IQ handles retrieval planning across knowledge bases and internal repositories. The fourth layer, Web IQ, covers everything outside the corporate perimeter — and it’s the one with reach well beyond Microsoft’s existing enterprise customer base.

Agents That Stay Running

Alongside IQ, Microsoft introduced the Foundry Agent Service, a managed hosting environment for production AI agents. The core design shift is statefulness.

Most AI interactions are stateless: a prompt arrives, a response leaves, the system retains nothing. That works for a single-turn exchange but breaks for enterprise tasks that span hours. A contract review running four hours can’t restart from scratch when a network connection drops. The Foundry Agent Service blog compares this inflection point to where microservices were a decade ago: one service is tractable to build, but the infrastructure around it (isolation, observability, deployment) is where the real engineering difficulty lives.

Each hosted agent runs in a VM-isolated sandbox with dedicated compute, memory, and a persistent filesystem. Agents survive container crashes, redeployments, and idle periods, resuming with state intact. The runtime accepts agents built on Microsoft’s own Agent Framework, LangGraph, the OpenAI Agents SDK, or the Anthropic Agent SDK, with no code rewrites required.

• Early July 2026: target general availability for hosted agents in Foundry Agent Service, per Microsoft Foundry’s Build 2026 release notes
• +7 to 14 percentage points: absolute success-rate improvement on Tau-bench, Microsoft’s published result from enabling procedural memory
• 5%: improvement on STATE-Bench, the open-source stateful agent evaluation benchmark, with procedural memory active

Procedural memory, now in public preview, captures which execution steps led to successful task completion across prior runs — not just what was said in a session. When a similar task arrives later, the stored procedure gets injected into the agent’s context, guiding it along a proven path. The agent optimizer, entering public preview this month, reads production traces, proposes ranked prompt improvements with full diffs, and surfaces an audit log and rollback path before any change is promoted.

The MAI Model Family

Build 2026 confirmed a strategic shift building since Microsoft hired Mustafa Suleyman, executive vice president and CEO of Microsoft AI, to lead first-party model development. The official Microsoft Build 2026 blog announced seven in-house MAI (Microsoft AI) models, developed by the Microsoft AI Superintelligence Team and trained from scratch on commercially licensed data with no distillation from third-party model outputs. The family spans the full enterprise compute spectrum:

• MAI-Thinking-1: First reasoning model, 35 billion active parameters, 256,000-token context window, in private preview on Foundry
• MAI-Code-1-Flash: Inference-efficient coding model, rolling out to GitHub Copilot and VS Code from June 2
• MAI-Image-2.5: Text-to-image and image-to-image workloads, live in PowerPoint and rolling out to OneDrive
• MAI-Image-2.5 Flash: Faster variant for lower-latency image generation
• MAI-Transcribe-1.5: Transcription across 43 languages, with streaming support forthcoming
• MAI-Voice-2: Natural speech in more than 15 languages, with voice-cloning safeguards and output watermarking built in
• MAI-Voice-2 Flash: Low-cost, low-latency variant built for real-time voice agent workloads

MAI-Thinking-1 is the flagship. Per the official Build blog, it scored 97% on AIME 2025, a mathematical reasoning benchmark, and matches Claude Opus 4.6 on SWE-Bench Pro, the software engineering coding benchmark. Independent rater firm Surge preferred it over Claude Sonnet 4.6 in blind evaluations. Microsoft projects roughly 10x more output tokens per dollar compared to GPT-5.5, citing serving-cost differentials scaled across model sizes. These are Microsoft’s own figures; independent benchmark verification has not been published.

Zero-distillation training carries real weight for regulated buyers. Banks, insurers, and health systems facing legal scrutiny over AI training-data lineage can build on MAI models without inheriting intellectual property exposure from another lab’s outputs. The full MAI family is also available on OpenRouter, Fireworks AI, and Baseten alongside Foundry. As covered on this site, Anthropic’s Claude Opus models run on Microsoft Foundry too, so the catalog now carries both Microsoft’s first-party models and its primary third-party alternatives inside the same procurement layer.

Web IQ and the Infrastructure Bet

Three of the four IQ layers operate inside Microsoft’s existing enterprise data estate: Work IQ on Microsoft 365, Fabric IQ on OneLake, Foundry IQ on internal repositories. Web IQ reaches the open internet without requiring any specific model on the developer’s side — and it already powers the web grounding inside both Microsoft Copilot and OpenAI’s ChatGPT.

According to Microsoft’s Web IQ product page, the service runs at 164ms p95 latency, “nearly 2.5x faster than today’s best alternative.” Microsoft’s benchmark tested against unnamed competitors labeled Competitor A through G, and the configurations may not reflect every real-world deployment scenario. The model-agnostic specification means a team running Claude, Llama, or any third-party reasoning engine can send retrieval queries to Web IQ and receive structured web passages in response, without adopting a Microsoft model. Bing’s global search infrastructure sits underneath, rebuilt from the ground up for AI-native workloads: indexing, retrieval, ranking, passage selection, and orchestration all redesigned for machines that consume results at inference time rather than humans scanning links.

The comparison point is the existing Bing Search API, available inside Azure. That API delivers raw search results and leaves the retrieval-augmented generation (RAG, the technique of supplementing model outputs with retrieved documents) pipeline to the developer — HTML parsing, text cleaning, content reranking, stale-result filtering. Web IQ packages that pipeline and returns passage-level evidence ready for direct injection into an agent’s context window.

Bing already provided web retrieval inside ChatGPT before this announcement, so the pattern of turning Microsoft’s search index into a revenue stream that flows through a different company’s product is established. Web IQ extends and formalizes that arrangement for the enterprise agent market: a managed retrieval service competing with the hours developers would spend building equivalent capability themselves. Jordi Ribas, Microsoft’s president of Search and AI, confirmed the ChatGPT integration in media interviews around the Build launch, though he declined to name additional future customers at that time.

Governance as the Differentiator

Neither Google’s Vertex AI Agent Builder nor Apple’s enterprise APIs wire directly into Microsoft Entra, Active Directory, Purview, and Defender. Those four components are what most large enterprise security stacks already run on, and Agent 365 connects AI agent identities into all of them simultaneously.

Agent 365, now generally available, expands at Build 2026 to govern agents running across AWS, GCP, and Azure in a single control plane. Each agent gets its own enterprise identity through Microsoft Entra (the identity and access management platform), with Purview and Defender as the unified compliance backbone. Every interaction produces an auditable record: which data sources the agent queried, which permissions it consumed, what actions it took. For finance and healthcare deployments, that audit trail is a regulatory requirement, and Microsoft builds it into the platform layer so development teams don’t have to add it separately.

Microsoft Execution Containers (MXC, policy-controlled agent sandboxes at the OS level) extend the same principle down to Windows itself, restricting which files, applications, and system resources an agent can reach. A bank deploying an MXC-bounded agent has a defined, auditable blast radius enforced by policy rather than prompt-level instructions.

Windows runs on over a billion machines; Microsoft 365 sits inside most large corporate IT budgets. A governance layer installed at that scale reaches enterprise agents faster than any greenfield deployment can. Models are commoditizing and retrieval APIs are converging. Security and compliance depth, built over two decades of enterprise IT deployment, is what takes longest to replicate.

The Work IQ APIs go live on June 16, the first opportunity for developers outside Microsoft to put IQ’s enterprise context promises under real production load.

DMZ launched November 16, 2022 as part of Warzone 2.0, ran under the “Beta” label through seasonal updates in 2023, and was retired that year. Infinity Ward described it as a successful beta. No Call of Duty title shipped the mode again in the three years since.

Abandoned After the Beta

The original DMZ’s run covered five exclusion zones: Al Mazrah, Building 21, Ashika Island, Koschei Complex, and Vondel. It drew players throughout its first year and received faction storylines, weekly missions, and seasonal updates. The problem, as PC Gamer described it, was that the mode was “casual, but bland,” and felt “more of an offshoot of Warzone than its own game.” Game Rant put the structural issue plainly: DMZ suffered from “a lack of structure, long-term progression, and ultimately a reason to keep logging into the mode.”

Infinity Ward didn’t label it “beta” ironically. Faction missions had no persistent progression outside the seasonal grind. Players built faction relationships, unlocked mission tiers, and earned gear, then watched that investment reset when each new season launched. By Season 4, mission trees had been restructured multiple times. A portion of dedicated players kept returning regardless.

Each seasonal update reset mission pools and faction access. A player who completed Season 2’s faction mission tree arrived at Season 3 with those missions wiped. The mode held its beta designation from launch to retirement, and both Black Ops 6 and Black Ops 7 shipped without it.

Seven features in the MW4 set each have a 2022 analogue that was absent or incomplete.

Hajin as a Living Exclusion Zone

The MW4 DMZ map is Hajin, a dense urban zone on the Korean Peninsula bordering both Russia and North Korea. It’s positioned after the events of MW4’s campaign, when a nuclear reactor meltdown leaves the area contested over abandoned military technology and weapons stockpiles. Twenty squads of up to three players deploy per match as off-the-books assets tasked with recovering that technology before hostile forces or rival operators claim it.

Rival factions, roaming patrols, and shifting world events ensure that no two operations unfold the same way.

Activision’s June 7 DMZ overview, released ahead of the Xbox Games Showcase, describes the zone operating on three categories of dynamic variables absent in the 2022 version. Weather changes per match: sudden downpours, rolling fog, and significant temperature drops alter sightlines, movement, and which objectives stay accessible. Environmental storytelling and hidden discoveries are built into the deeper parts of Hajin, with the overview calling them rewards for players willing to push past early extraction points.

Hajin’s design philosophy places those hidden discoveries and environmental details in areas beyond the immediate deployment zones. Players who push deeper encounter narrative context and richer loot that closer extraction points don’t carry. The zone’s density is described as built around exploration, finding what previous operations left behind in a space designed to be revisited across multiple runs.

Escalating threat levels tie time-in-zone directly to enemy response. Early in a deployment, forces are standard patrols. As a squad pushes deeper and stays longer, responses become more coordinated, eventually triggering elite units specifically dispatched to hunt the squad’s position. Dynamic Operations run alongside this structure as randomly generated per-match missions with evolving scenarios and a unique climactic moment each run, separate from the scripted Story Missions. A squad chooses how structured or freeform its deployment will be within that framework.

Persistent Progression and the FOB

The Hub Between Deployments

The Forward Operating Base, detailed in the MW4 extraction mode overview on the official Call of Duty site, functions as the persistent hub between runs. Players upgrade it to unlock new services and expand operational capabilities before their next deployment. Vendors, loadout setup, and crew management live inside the FOB, and the hub retains value across sessions rather than resetting each time a player logs off.

Players who develop the FOB gain access to expanded vendor inventory, additional crafting options, and more operator slots. That infrastructure was absent in the 2022 version, where players jumped from the main menu directly into the exclusion zone with no persistent home base to return to and no upgrade path to pursue between sessions.

Operators recruited through DMZ sessions carry their own progression tracked across the mode. They can be customized between runs, and their narrative context connects to the broader story structure Infinity Ward built around the campaign. The FOB also handles incoming gear: weapons, materials, and equipment flow back from Hajin into a persistent inventory.

Crafting and Persistent Inventory

The 3D Printer system produces equipment, weapons, and high-end gear from materials gathered in Hajin. Damaged weapons recovered in the field have their own repair path, and rarer enemy-variant weapons found on AI combatants carry a separate recovery process. Failed extractions impose real consequences on inventory, but the mode description states recovery features keep players in progression even after a run ends badly.

The Bounty Board adds a competitive layer on top of the PvE structure. Killing rival Operators builds notoriety and raises a player on a public leaderboard, surfacing them as a high-value target to other squads. Per the Modern Warfare 4 first-look coverage on Xbox Wire, every PvP cross-squad encounter inside DMZ carries leaderboard stakes attached to the notoriety earned.

• Forward Operating Base (FOB): Upgradeable persistent hub with vendors, services, and loadout management between sessions
• Story Missions: Narrative objectives tied directly to the MW4 campaign, built for squad replayability with varying outcomes
• Dynamic Operations: Randomly generated per-match scenarios with evolving stakes and a unique climactic moment each run
• 3D Printer: Crafts weapons, equipment, and high-tier gear from materials collected inside Hajin
• Bounty Board: Tracks rival Operator kills, builds notoriety, and feeds a competitive leaderboard turning PvP engagements into reputational stakes
• Escalating AI Threats: Enemy response strength scales with time in-zone, with elite hunter units dispatched at peak threat levels

Story Missions Tied to the Campaign

MW4’s campaign centers on Private Park, a South Korean soldier played by Young Mazino (known for the series Beef and The Last of Us), navigating an invasion of the Korean Peninsula. The DMZ’s Hajin zone sits after those campaign events: a nuclear reactor meltdown leaves the area as a contested exclusion zone, with the abandoned weapons and military technology littering the map being the direct consequence of the campaign’s outcome.

Private Park’s story and Hajin’s mission set share the same geography and timeline in the fiction. Story Missions inside DMZ play out as operations a player could imagine the broader task force handling in the campaign’s aftermath. Activision describes them as “built for replayability, squad-based action, and evolving storytelling,” with outcomes that vary based on squad composition and decisions across replays.

Missions sit alongside Dynamic Operations in the structure, giving squads the choice of a scripted narrative run, a randomized operation, or a freeform deployment through Hajin with no fixed objective. The faction mission system in the 2022 Beta had no connection to Modern Warfare II’s campaign. Players ran faction objectives inside Al Mazrah, but those had no bearing on the game’s main story and no narrative thread beyond faction flavor text. The mode description for MW4 specifically frames Story Missions as purpose-built to create continuity between the campaign and the extraction mode, with the same events and geography tying both together.

Into a Crowded Extraction Market

When DMZ launched in 2022, Escape from Tarkov had dominant mindshare in the extraction category and the genre was largely understood as a hardcore niche. ARC Raiders and Bungie’s Marathon are both shipping into the same space in 2026, competing for a player base the genre has built across those four years. Hunt: Showdown, now in its second iteration, has also continued developing the middle market since DMZ’s 2022 run. Between those titles and several smaller entries, the extraction format has gone from a category with two active choices to a crowded field by mid-decade.

Activision bills the mode as “the definitive Call of Duty extraction experience” in the June 7 overview. DMZ’s relaunch enters that field with the widest platform footprint in the genre: PS5, Xbox Series X|S, PC via Steam, Battle.net, and the Microsoft Store, and Nintendo Switch 2, all on day one. The 20 squads of up to three players per match structure sits at the more tactical end of the extraction format, distinguishing DMZ from battle-royale-scale player counts. The Nintendo Switch 2 version is developed by Digital Legends Entertainment, making MW4 the first Call of Duty on Nintendo hardware since Ghosts appeared on the Wii U in 2013.

The Price and the Game Pass Reversal

MW4 launches at $70 for the standard edition and $100 for the Vault Edition, which includes operator skins, weapon blueprints, and a BlackCell Season Pass. Players who have owned and played a qualifying Call of Duty title since 2019 get 10% off the Vault Edition on pre-order, per the terms listed on the Xbox Store’s Modern Warfare 4 pre-order page.

Xbox Game Pass subscribers won’t get it at launch. An April 2026 policy shift confirmed that MW4 will not follow Black Ops 6 and Black Ops 7 onto the service on release day. MW4 is expected to join Game Pass around late 2027, roughly 12 months after launch, as detailed in the Game Pass policy change affecting Modern Warfare 4. That wait spans at least one full annual CoD cycle.

The commercial context behind the policy shift shows in Steam data: Black Ops 7 sold approximately 401,000 copies in its first 26 days on Steam, against Black Ops 6’s 2.3 million, per Gamermarkt analysis cited by Gagadget. MW4 exits that day-one subscription model entirely. It also launches 27 days ahead of GTA 6, which Rockstar is targeting for November 19, giving Activision a clear sales window before the year’s other major release arrives.

Warzone’s removal from Xbox One and PS4 on June 4 ended the franchise’s free-to-play extraction presence on last-gen hardware just days before the DMZ overview landed. The consolidation around current-gen places the entire CoD extraction ecosystem on the same hardware as MW4 ahead of October.

DMZ launches alongside the campaign and multiplayer on October 23, on every supported platform simultaneously.

The game arrives on console after 14 months in Steam Early Access, during which it sold over one million copies and built a Very Positive rating from more than 19,000 reviews. Xbox players get the full 1.0 build, complete with new content shipping the same day.

RuneScape’s First Console Home

The launch is the first time any title in the RuneScape franchise has appeared on a console, per Jagex’s official full-launch announcement. RuneScape began as a browser-based MMORPG (massively multiplayer online role-playing game) in 2001, and the franchise has accumulated over 300 million lifetime player accounts. The scale places RuneScape among the most-played PC and browser games in gaming history, a franchise that remained continuously active for 25 years without ever shipping a console title.

The gap follows from the original game’s design. RuneScape ran in a web browser, requiring only a Java plugin and an internet connection, which made it one of the most accessible games of the early 2000s and anchored its identity firmly in the PC ecosystem. The browser-based point-and-click control scheme was also poorly suited to a gamepad, and Jagex never built a console version of either MMO. As the franchise grew, it expanded to mobile and split into two distinct live versions: RuneScape 3, which modernized the graphics and systems progressively, and OldSchool RuneScape, a preserved 2007 snapshot that has consistently drawn the larger concurrent player base of the two. Both remain active on PC and mobile today. Neither has appeared on Xbox or PlayStation. Dragonwilds, a third-person action game built on Unreal Engine 5, is the first RuneScape product designed around a controller.

Jagex describes 2026 as the franchise’s 25th year, and Dragonwilds is positioned as the opening move in a strategy to build what the studio’s launch press release calls “a broader ecosystem of forever games and connected experiences.” Dragonwilds is the first consumer product from that plan.

• September 15, 2026 – full 1.0 launch across PC, PS5, and Xbox Series X|S simultaneously
• 25 years – the RuneScape franchise’s history, from the 2001 browser game forward
• 300 million+ – lifetime player accounts across RuneScape’s MMO titles
• 1 million+ – copies sold by Dragonwilds during its Steam Early Access period

Survival Crafting in the RuneScape Universe

A New Genre Built on Familiar Lore

Dragonwilds is a co-op survival crafting game, sharing almost nothing mechanically with the RuneScape MMOs beyond brand names, skill labels, and lore. The genre sits closer to Valheim than to the point-and-click MMORPG most players associate with the RuneScape name. Up to four players explore Ashenfall, a continent in the RuneScape universe that has never appeared in the MMOs, gathering materials, building shelters, leveling skills, and fighting progressively harder enemies on the way to a Dragon Queen boss encounter.

The game runs on Unreal Engine 5 (UE5, Epic Games’ current-generation rendering platform) and uses magic as its core mechanical hook. Players can summon a spectral axe to harvest trees instantly, teleport out of combat with a blink ability, and draw on the continent’s ambient magical energy (the game calls it anima) to speed up tasks that are purely manual in other survival titles. Early reviewers in 2025 consistently cited the magic system as the element that most clearly distinguishes Dragonwilds from the Valheim comparison that coverage reached for at launch.

The game supports solo play and co-op in teams of up to four players, with progression tied to individual characters. Joining a friend’s world mid-session carries the player’s existing skill levels and stats into that session. That design sidesteps a common friction point in co-op survival games, where a player who joins a world their teammates have developed for weeks must grind through early-stage content the host has long since cleared.

Ashenfall and the Path to the Dragon Queen

Jagex built Ashenfall as a hand-crafted world rather than a procedurally generated map, a decision that distinguishes it from the random-terrain approach common in the genre. Locations unlock in sequence as players clear bosses and advance skills, following a quest structure familiar to RuneScape players. The central storyline takes its name from Dragon Slayer, one of the most celebrated quests in the original MMO, and leads ultimately to Kuldra the God Eater, the Dragon Queen who serves as the game’s final boss.

The RuneScape lineage appears in the enemy list and the character roster. Zogres, Black Knights, and a new creature type called the Garou populate Ashenfall. The Wise Old Man, the franchise’s most recognizable NPC (non-player character), sets the player’s quest in motion on arrival. New players encounter these details as franchise flavor. Longtime RuneScape players find them loaded with two decades of association, without the game needing to flag the references.

Xbox Play Anywhere Covers Both Versions

When Dragonwilds sells on Xbox, it sells as an Xbox Play Anywhere title. The program entitles the buyer to both the console and Windows PC version from a single Xbox Store purchase, with cloud saves syncing between platforms and Xbox Achievements shared across both. Save progress on Xbox and continue on a Windows machine, or start on PC and switch to the console, without replaying anything.

Dragonwilds spent more than a year building its audience on Steam before the Xbox version was announced. A co-op survival crafting game depends on group formation, and Play Anywhere removes one barrier: an Xbox buyer can join sessions with friends running the Windows version through the Microsoft Store, carrying the same character progression. In survival crafting games, that progression represents significant time investment, skill levels accumulated over hours, gear upgraded, crafting recipes unlocked. The cloud-save sync means that investment transfers between platforms. For a game whose Steam community has been active since April 2025, the console launch extends that existing base rather than splitting it by platform.

For titles in Microsoft’s cross-platform purchase program, the entitlement generally also covers Xbox Cloud Gaming (Microsoft’s game-streaming service), allowing the game to be played via browser or supported device without a console or gaming PC installed. The Microsoft Xbox Wire announcement for Dragonwilds confirmed cross-platform purchase support as part of the Xbox launch package.

• One purchase covers Xbox and PC: a single Xbox Store transaction includes Xbox Series X|S and the Windows PC version via the Microsoft Store
• Cloud saves sync automatically, carrying progress between console and PC
• Xbox Achievements are unified across both versions

Fourteen Months in Early Access

One Million Sales and a Very Positive Rating

Dragonwilds entered Steam Early Access on April 15, 2025, priced at $29.99 USD. By Summer Game Fest 2026, Jagex confirmed the game had surpassed one million paid copies during that period. More than 19,000 Steam reviews gave it a Very Positive rating, with 81% of reviews marked positive as of the announcement week.

Early reviews in the first months praised the building system, the skill progression loop, and the magic-based resource gathering. Recurring criticisms hit inventory weight limits that felt overly tight, loot tables that generated cosmetic recipes before useful gear, and combat that reviewers described as functional but lacking physical impact. Jagex acknowledged the feedback publicly and structured the update schedule around those specific complaints. The rating improved as the patches arrived.

The Updates That Built Version 1.0

The development roadmap was public from Early Access day one, and Jagex delivered major content drops on roughly the cadence it outlined. In September 2025, version 0.9 added the Ranged skill, widening combat options beyond the melee-and-magic setup players had used since the April launch. Three months later, in December 2025, the Fellhollow update introduced a new realm built around decay and death, added Imaru (a corrupted guardian dragon) as a boss encounter, brought in magic-based farming, and added Terrorbird riding as a traversal option. Each major update carried feature additions directly tied to player feedback, per Jagex’s stated development approach throughout the Early Access window.

Between April 2025 and the Summer Game Fest reveal, the game grew from a build reviewers called promising but incomplete to a version the studio felt confident launching on three platforms simultaneously. The update cadence kept the player base engaged across 14 months before the 1.0 launch.

Our community has been at the heart of RuneScape: Dragonwilds from day one. Every update throughout Early Access has been shaped by player feedback, and we’re incredibly proud of what we’ve built together.

Jesse America, Executive Producer of RuneScape: Dragonwilds, in the RuneScape.com launch announcement, June 5, 2026.

New Content at Full Release

The 1.0 build ships with two content packages not included in any prior release. The Scorned Wilderness update is new to version 1.0, arriving across all platforms on September 15. The Umbral Sands expansion is set for PC during Summer 2026 ahead of the full launch date, so it will be built into the complete package console players receive from day one.

PC players also get a new distribution channel at 1.0: the Epic Games Store version releases simultaneously with Steam, the first time Dragonwilds has been available outside Valve’s platform. The rollout across Xbox Series X|S, PS5, Steam, and the Epic Games Store is the first all-platform simultaneous launch for the game; prior to this, only Steam players had any version.

Players can wishlist the Xbox version on the Xbox Store now. The listing went live immediately following the Summer Game Fest reveal on June 5.

• The Scorned Wilderness update, new to version 1.0
• All Umbral Sands expansion content, integrated before launch day
• All prior content from the Early Access period, including the Fellhollow realm, the Ranged skill, and Terrorbird traversal
• Solo and co-op play for 1 to 4 players

Mid-September delivers the franchise’s first console game in 25 years, and the Xbox version carries the dual-platform entitlement from day one.

The campaign traces directly to the March 2022 collapse of the Conti ransomware syndicate, which left behind a callback-phishing unit experienced at convincing corporate employees to hand over remote access. SRG inherited that technique, dropped encryption, and built an extortion model around the sector with the most to lose from public data exposure.

How a Fake IT Call Becomes a Data Breach

The campaign opens with an invoice or billing-themed phishing email, sent from a consumer email account and carrying no malicious links or attachments. Its purpose is purely preparatory: manufacture a plausible reason for a follow-up call.

From there, Mandiant documented a consistent attack sequence across dozens of incidents from January to May 2026:

1. Invoice phishing email arrives clean from a consumer account, directing the recipient to call a provided number to dispute or cancel a charge.
2. The call connects to an attacker posing as the firm’s own IT help desk, citing a pending data migration, a security scan, or follow-up from the phishing email itself as justification.
3. The target is invited to join a remote session via Microsoft Teams, Zoom, Quick Assist, or Microsoft Terminal Services, presented as standard troubleshooting.
4. During the session, the attacker directs installation of legitimate remote monitoring and management (RMM) tools including AnyDesk, Zoho Assist, Bomgar, or SuperOps, establishing persistent access without triggering security alerts.
5. Inside the environment, operators search for contracts, tax records, Social Security numbers, merger and acquisition files, and materials stored on cloud repositories and document management platforms.
6. Data exits the network via WinSCP (Windows Secure Copy) or Rclone, a file-transfer utility the group often renames or hides to reduce detection.

To reduce forensic traces, operators use privnote[.]com, a self-destructing messaging service, to send installation commands during remote sessions so the links don’t linger in browser histories or corporate chat logs. Mandiant also documented phishing domains built on patterns like [organization]-itdesk[.]com and [organization]-helpdesk[.]com, designed to mimic internal support addresses that employees wouldn’t question.

In one confirmed intrusion, a single target took five separate Teams calls over three days as the attacker incrementally expanded remote access. In more than one investigated case, attackers established Zoom sessions on employees’ personal laptops, then used those machines to reach corporate virtual desktop infrastructure via Windows 365 or Citrix clients.

Why Law Firms Are the Primary Target

SRG has focused on U.S. law firms since at least spring 2023. In its June 2026 threat intelligence report on the UNC3753 campaign, Mandiant cited law firms’ “concentrated repositories of extremely sensitive client transaction files” as the primary attraction, noting that their heavy reputational and regulatory exposure motivates quiet resolution over public disclosure. Law firms also store attorney-client privileged communications, M&A documentation, and intellectual property records, giving SRG leverage that is difficult to quantify and hard to neutralize by paying.

The consequences for firms that declined to pay are now on the public record. In January 2026, Orrick, Herrington & Sutcliffe, an international firm with more than $1.5 billion in annual revenue, had its data posted publicly after declining SRG’s demand. Jones Day and Wood Smith Henning & Berman each faced similar exposure in the first quarter. Wood Smith countered SRG’s $1.8 million demand with an offer of $15,000; the group posted the files. As recently as May 6, SRG claimed a breach at Ropers Majeski, a California litigation firm.

Halcyon, a cybersecurity firm, tracked 134 incidents against law firms in Q1 2026 alone, placing legal as the fourth-most targeted industry at more than 6% of all tracked ransomware-related incidents that quarter. SRG and the INC ransomware-as-a-service operation were identified as the primary drivers of that increase.

“The theft of data in and of itself is the biggest issue for the law firms,” said Cynthia Kaiser, senior vice president at Halcyon’s Ransomware Research Center. “They’re tailoring a lot of their operations around what they know about the sector.”

Out of Conti’s Collapse, a Leaner Model

SRG’s direct ancestor is UNC2686, a Mandiant-tracked threat cluster that ran BazarCall-style callback phishing campaigns from early 2021, supplying initial network access to Conti and Ryuk ransomware operators. When Conti disbanded in March 2022, the people behind UNC2686 didn’t retire. They reorganized as Silent Ransom Group, briefly deployed LockBit.Black ransomware that year, and abandoned encryption-based attacks entirely in favor of data-theft extortion.

The group’s earliest lure was a subscription billing email, instructing recipients to call a number to cancel a pending charge. Those calls walked employees into downloading remote access software, laying the operational foundation that the current IT-impersonation approach refined.

The group scaled callback phishing against a broad target set in 2022. Through 2023 and 2024, it industrialized around legitimate RMM tooling and rapid exfiltration, concentrating increasingly on law firms. In 2025, operators shifted toward direct vishing with IT impersonation and lookalike helpdesk domains. The Spring 2026 escalation to physical access followed when remote social engineering started meeting organized employee resistance.

The FBI has made no arrests, noting SRG’s operators are believed to be Russia-based. To carry out English-language vishing calls and in-person visits, researchers told CyberScoop, the group likely employs local gig workers or subcontractors who may not fully understand they’re facilitating a crime.

When the Phone Fails, Knock on the Door

The in-person physical tactic prompted the FBI to issue FLASH-20260526-01 on May 26, its second formal warning about SRG in twelve months and its first at FLASH severity, a classification reserved for active threats requiring immediate action from organizations in the affected sector.

If that attempt fails, SRG sends a threat actor to the victim’s location to gain access to insert a storage device into the victim’s computer.

The FBI described the operative’s cover story: they tell the victim they need to image the device or create a backup file to address potential impacts from the earlier phishing email. Once on-site, the operative connects a USB drive or external hard drive to a workstation, exfiltrates data, and leaves. The FBI noted SRG escalates privileges minimally; the goal is speed, not depth.

Ian Gray, vice president of cyber threat intelligence operations at Flashpoint, a threat intelligence company, called the tactic “extraordinary,” noting that while some extortion groups have threatened employees with physical harm, “we have not observed them physically sending attackers to victim locations.” With SRG’s core operators Russia-based, researchers told CyberScoop the in-person operatives are likely local freelancers recruited through online channels.

Extortion demands arrive fast. Mandiant documented ransom emails appearing within 30 minutes of attackers leaving the victim environment. The three-day deadline SRG sets in its letters gives firms almost no time to assess their exposure before the group threatens to call client contacts directly to announce the breach.

The Botnet Keeping SRG’s Leak Site Online

In a companion report published days after the FBI FLASH alert, Resecurity, a cybersecurity firm, documented the fast-flux botnet infrastructure SRG uses to protect its data leak site from takedowns. The technique cycles a domain’s IP addresses continuously through a pool of compromised residential devices, primarily home routers and connected hardware across multiple regions, making blocking attempts expensive and slow. Fast-flux infrastructure significantly raises the cost of defensive countermeasures, since standard IP-based blocking loses effectiveness against a pool of thousands of rotating residential addresses.

Unlike most ransomware operators who host leak sites on the Tor network, SRG runs its site on the public internet. Resecurity’s analysis noted the clearnet choice lowers the barrier for victims and journalists to access posted data, amplifying reputational pressure. All domain names associated with the infrastructure remained fully operational as of June 5, 2026. The firm also linked the network to a separate underground project called Spy Corporate, which emerged in May 2026 and shares IP addresses with the original SRG infrastructure.

The site lists each published victim with their revenue and the download count for their stolen data, framing the exfiltrated files as browsable corporate intelligence for competitors and adversaries. New entries were added as recently as the first week of June 2026.

• 100+ law firm and professional services attacks claimed since early 2023
• $1 million to $8 million typical extortion demand, scaled to firm revenue
• ~100 victim organizations listed on business-data-leaks[.]com as of early June 2026
• 20+ countries hosting fast-flux botnet nodes across Latin America, Eastern Europe, Central Asia, and East Asia

Stopping Attacks That Look Like Your Own IT Team

Traditional antivirus products are unlikely to flag an SRG intrusion because the group uses legitimate system management and remote access tools throughout. No unusual executables. No malware signatures. The FBI confirmed the same in FLASH-20260526-01: the tools SRG installs are indistinguishable from what a real IT team would deploy.

Mandiant and the FBI recommend building defenses across both digital and physical layers, recognizing that SRG’s playbook exploits gaps in both:

• Establish and communicate a formal procedure for how IT contacts employees; unsolicited calls that don’t follow that procedure should be treated as suspicious by default.
• Require independent callback verification through a known internal number before granting any remote desktop session to an unscheduled caller.
• Audit and restrict which RMM tools are permitted on corporate endpoints, and flag unauthorized installations of AnyDesk, Zoho Assist, Bomgar, or SuperOps.
• Enforce multifactor authentication (MFA) across all employee accounts, particularly those with access to document management systems and cloud repositories.
• Block or restrict USB storage device access on workstations, requiring IT authorization before external media can be connected.
• Apply physical security verification at reception: require identification from anyone claiming to be an IT support visitor, and cross-check against the firm’s own schedule.
• Monitor outbound WinSCP and Rclone connections to external IP addresses as an indicator of possible exfiltration in progress.

A joint advisory from CISA and security agencies in Australia, Canada, New Zealand, and the UK designated fast-flux DNS as a national security concern and called on ISPs and DNS providers to implement detection and blocking measures at the network level.

SRG has told researchers that most of the firms it attacks do pay quietly; the more than 38 whose data landed on its clearnet leak site are the ones that refused.

The document was co-authored by Omar Shahine, the corporate vice president leading the Scout project, and by a second executive, Jakob Werner. CEO Satya Nadella has since rebuked it publicly, calling the document “nonsense,” but the authorship question he raised is easier to resolve than his message implied, and the publication he rebuked has pushed back firmly on his framing.

The Three-Phase Plan

The Agent at Launch

Microsoft had been developing the agent under the codename ClawPilot, as part of a project called Project Lobster, long before the June 2 announcement. The product runs on OpenClaw, the open-source agentic framework created by developer Peter Steinberger, which accumulated 180,000 GitHub stars in the three months after its November 2025 launch. Scout connects to Teams, Outlook, OneDrive, and SharePoint, carries its own Entra identity, and acts continuously on a user’s behalf without waiting for a prompt; unlike a standard chat assistant, which responds only when asked.

The internal pilot showed the concept working before any public launch:

• 1,000+ Microsoft employees, including the CEO, used the agent internally before the public announcement
• 20 million paid Microsoft 365 Copilot subscribers as of May 2026, the installed base the agent is designed to expand into
• Microsoft targets October 2026 for the agent’s general availability
• 3 phases in the strategy document, describing a journey “from addictive app to agentic platform”

Phase One’s Stated Goal

The strategy document, titled “ClawPilot: Overview and Plan with Project Lobster,” described that journey in three phases. Phase one’s heading stated: “Make people addicted.” The instructions beneath told the team to “continue shipping the standalone ClawPilot experience, pilot the UX, grow the user base, and build the skill and tool ecosystem that makes people depend on it daily,” per 404 Media’s reporting. The document added that the process was “already happening organically.”

Internal pilot data backed that claim. Employee testers showed what the document described as “Daily Usage with High Retention and intensity of usage (chats, queries, workflows, skills),” and the agent had spread widely inside Microsoft without a formal announcement, no marketing push, and no org-wide directive. The document cited that organic spread as evidence the approach would translate externally.

Phase two covers linking the platform to broader AI networks; phase three positions the agent as a full agentic ecosystem open to third-party developers. Microsoft announced an SDK at the conference enabling external developers to build custom skills into the agent, which maps directly to the “agentic platform” endpoint in the strategy document.

One anonymous Microsoft employee told the publication the language was “very troubling,” saying addiction “is something no product should be making a part of its build strategy.”

The VP Who Wrote Both

As corporate vice president for the Scout project, Shahine wrote the official launch blog post introducing the agent publicly and was quoted in coverage describing Autopilots as systems that act autonomously “without needing to be prompted each time.” His role as the project’s lead is documented on his personal blog, his LinkedIn page, and in Microsoft’s own press materials for the event.

His name also appears as a co-author on the leaked strategy document, alongside Werner. Together they produced two documents in the same week describing the same product in notably different terms.

At the conference, Microsoft demonstrated the agent preparing an entire quarterly business review without the user switching between a single app, per Bloomberg’s coverage of the event. His public blog post described the same product in terms of autonomous task completion, risk surfacing, and coordination, with no mention of dependency or retention as goals.

Consumer tech companies routinely target daily engagement and habit formation in product strategy. “Addicted” is a step further than the industry’s standard language of stickiness or daily active users. The document also carried a note that it was “co-created turn-by-turn with AI. Human verified every sentence.” That attribution doesn’t change the two authors’ formal ownership; both are named in the document, and their identities as the project’s leads are publicly documented.

A Rebuke With a Fuzzy Target

What Nadella Wrote

Nadella’s response, sent to roughly 50 of Microsoft’s top engineers working on AI products and reported by The Information, came after the document’s public exposure. He attached a copy of the report and posted the following to an internal board:

This is absolutely a non goal! If anything we are doing the exact opposite. We want to make sure AI empowers and adds real value to human endeavor and broad economic growth! We should make sure that our teams are clear about this. Not sure what this document is or who is writing and leaking this nonsense! They may want to go work elsewhere.

The rebuke went to The Information. So did Microsoft’s prepared statement on the subject. Frank Shaw, a Microsoft spokesman, told the outlet that the agent is designed to help people complete tasks more effectively, with the company’s goal being to give time back to users. Users will have “clear choice and control,” Shaw said.

Neither the rebuke nor the statement was shared with the publication that obtained and published the strategy document.

The Publisher’s Rebuttal

404 Media responded publicly, rejecting the framing that the document’s authorship was unclear. “The document we reported on was not some random document,” the outlet wrote. “As we wrote at the time, the strategy document was written by Microsoft executives Omar Shahine, Jakob Werner, and some sort of AI writing tool.” The outlet noted that the project lead’s identity is documented on his own blog, his LinkedIn profile, and in Microsoft’s own announcement materials, and that he had written publicly about his role on the project numerous times.

The outlet also described what happened before publication. It had specifically told Microsoft it was writing about the addiction language in the document and had asked for comment, context, and more information. Microsoft sent back a link to the project lead’s public blog post. The company then, per the outlet, “attacked our report internally and externally to another media outlet.”

The project lead who authored both documents remained the corporate vice president overseeing the agent through the week of reporting. Nadella’s internal message questioned the document’s origin while the document was publicly attributed to the executive who wrote the conference’s official announcement the same day.

When “Addicted” Became a Liability

On March 25, 2026, ten weeks before the strategy document became public, a Los Angeles County Superior Court jury found Meta and YouTube liable for addictive platform design, awarding $6 million in the first bellwether trial of consolidated litigation now involving more than 2,400 cases against social media companies. Meta was found 70 percent responsible. A second user trial is scheduled for July 27; a school district case begins June 15.

The litigation has drawn direct comparisons to the 1990s tobacco lawsuits, which forced the industry to restructure its marketing practices and settle for hundreds of billions of dollars. In those cases, the central question became what companies knew about addiction and when they knew it. A strategy document with a labeled phase headed “Make people addicted” sits close to that question, even if the agent’s users are enterprise workers operating inside organizational policy controls rather than teenagers on a social feed.

European law adds a separate dimension. Article 5 of the EU AI Act prohibits AI systems that deploy “subliminal, manipulative, or deceptive techniques” to materially distort behavior in ways causing or likely to cause significant harm. The prohibition has been enforceable since February 2025, with fines reaching €35 million or 7 percent of global annual turnover. The leaked document does not automatically place the agent under that prohibition; the EU Act targets behavioral manipulation causing harm, and the agent operates within organizational tenant-level governance. But the phrase is now visible to regulators in a way it wouldn’t have been without the leak.

What Enterprise Buyers Are Now Asking

The agent’s immediate market is enterprise organizations running Microsoft 365 E3 or E5. Getting access requires:

• A Microsoft 365 Copilot license
• A GitHub Copilot license
• Enrollment in Microsoft’s Frontier program, with Intune policy configuration and an opt-in attestation completed

That procurement gate means every adoption decision goes through a buying cycle that includes legal, compliance, and security reviews. Enterprise buyers don’t just read press releases; they assess what product teams appear to be optimizing for.

Microsoft 365 Copilot’s own adoption curve gives context. Around 3 percent of Microsoft 365 customers were paying for the Copilot add-on at the start of 2026, a figure that climbed to 20 million paid subscribers by May after more than a year of sustained selling. The agent enters that same enterprise base as a further add-on, with broader preview access opening in late June before the general availability push. At the same event, Microsoft unveiled seven new in-house MAI models, the first output from its own AI superintelligence lab, in a week meant to show enterprise customers a coherent AI roadmap. The strategy document became the story from that week that required the most explanation.

By October 2026, when the agent reaches general availability, its behavior inside enterprise tenants will have replaced the leaked document as the primary evidence of its design intent.

Google’s On-Device Dictation Reaches the Mac

Google AI Edge Eloquent, free and available from the Mac App Store, transcribes speech, strips filler words, and applies editorial polish through a customizable keyboard shortcut that works on top of any open application. All audio stays on the device. No internet connection required. Google released it alongside Gemma 4 12B on June 3.

The Gemma 4 12B Foundation

The model requires at least 16GB of unified memory and runs on Apple Silicon through Google’s LiteRT-LM inference engine. Google describes Gemma 4 12B as capable of performance comparable to its 26-billion-parameter mixture-of-experts variant while fitting the hardware limits of a standard consumer MacBook. It handles text, vision, and audio from a single checkpoint. Weights are on Hugging Face and Kaggle under Apache 2.0, permitting commercial use; companies can build products on top of the model weights without a separate licensing arrangement with Google.

Google’s LiteRT-LM command-line tool also gained a serve command this week, exposing a local endpoint compatible with the OpenAI API format and letting existing developer tools point at an on-device model without code changes. The companion app for running the Gemma family locally, Google AI Edge Gallery, limits users to Google’s own model weights as of this week’s launch. Platforms like Ollama and LM Studio accept any compatible open-weight model from Hugging Face; Gallery currently does not.

What Eloquent Handles and Its Limits

Voice Edit is the headline capability. Select text on screen, speak a command aloud, and the model rewrites it locally: translate a paragraph, summarize an email thread, reformat rough notes into an executive briefing. Output appears in whatever application is active. Custom vocabulary can be added for recurring proper nouns and technical terms to cut correction frequency, and writing style options let users match the tone of the target document.

The app launches English-only. Additional languages are promised but undated. On machines with less than 16GB of unified memory, users can fall back to lighter Gemma 4 E2B or E4B variants through the Gallery app, which handle simpler transcription tasks but lose the complex rewrite capability that Voice Edit requires.

Seven New Models at Microsoft Build 2026

On June 2, Microsoft announced seven in-house AI models spanning transcription, voice, image, reasoning, and coding, all trained on clean licensed data with zero distillation from third-party models and co-designed with the company’s proprietary Maia 200 AI accelerator chip. The family is available through Azure AI Foundry and the MAI Playground, with MAI-Thinking-1 still limited to private preview.

MAI-Transcribe-1.5 in Production

The model lands third on the Artificial Analysis speech-to-text leaderboard, a research firm that benchmarks commercial STT models, with a word error rate of 2.4% on the FLEURS multilingual benchmark across 43 languages. Alibaba’s Fun-Realtime-ASR-preview holds first at 1.7% WER; ElevenLabs’ Scribe v2 is second at 2.2%. The previous version, MAI-Transcribe-1, launched in April 2026 with 25 languages at the same price point; the 1.5 update expanded to 43 languages without raising the fee.

Speed is where the numbers pull away from the competition. Artificial Analysis clocks the model at roughly 276 times real-time, more than double the second-fastest model in the top ten for accuracy. Keyword biasing, detailed in Microsoft’s MAI-Transcribe-1.5 launch announcement, lets developers supply up to 200 domain-specific terms and cuts word error rate on specialized vocabulary by 30%. Pricing is $0.36 per hour of audio through Azure Speech. The model now processes audio behind Teams meeting transcription, Copilot, GitHub, and Dynamics 365 Contact Center, so Azure enterprise customers encounter it without a separate integration step.

Voice and the Broader Stack

All seven MAI models were built without data from OpenAI, Anthropic, or other labs. Mustafa Suleyman, Microsoft AI’s chief executive, framed the direction at the June 2 Build keynote:

building a hill-climbing machine to keep you at the frontier

MAI-Voice-2, the companion text-to-speech model, covers 15 languages with emotion and style controls and supports zero-shot voice cloning from as little as five seconds of sample audio, with consent safeguards built into the pipeline. Microsoft reports it was preferred over MAI-Voice-1 in 72% of side-by-side listener evaluations. Pricing is $22 per million characters through Azure Speech. The full family, available through Azure AI Foundry:

MAI-Thinking-1, a sparse mixture-of-experts model with 35 billion active parameters and a 256K token context window, matches Claude Opus 4.6 on SWE-Bench Pro and scored 97% on AIME 2025, per Microsoft’s published benchmarks. It remains in private preview pending wider production validation.

First-Look Tests Complicate the Benchmark Story

PCMag UK’s first-look comparison of Microsoft’s transcription model against Google’s Gemini cloud transcription model recorded 13 errors for the Microsoft model and six for Google’s in a real-world dictation session. The same first look described MAI-Voice-2 as sounding “robotic.” Both models are in limited preview, and Microsoft has not set a general availability date for either.

Both models in that session were cloud APIs. The result measures how the Microsoft transcription API performed against Google’s Gemini cloud model on specific test audio, not cloud against on-device processing.

The gap between leaderboard rank and a journalist’s dictation session has a structural component. FLEURS and the Artificial Analysis WER metric use curated multilingual audio under controlled conditions. A real-world session adds conversational pacing, ambient noise, and accent variation that benchmarks weight differently. Microsoft’s speed claim of 276 times real-time comes from batch transcription of pre-recorded clean audio, the workload the model’s current optimization targets most precisely. Microsoft’s preview label for both speech models acknowledges that production edge cases take longer to surface than benchmark runs.

Eloquent’s accuracy profile is shaped by different constraints. Running on a 12B-parameter on-device model means far less compute than a cloud API applies to the same audio clip. Automatic filler-word removal and editorial polishing help the output read cleanly, but for verbatim legal or medical dictation, where the precise wording of the original speech matters, that cleanup rewrites content the user intended to preserve.

Formal accuracy leadership sits outside both companies. Alibaba’s Fun-Realtime-ASR-preview holds first on the Artificial Analysis leaderboard at 1.7% WER, a figure neither company’s model has matched on that benchmark.

NVIDIA and Others Press the Open-Weights Case

NVIDIA released Nemotron 3.5 ASR on June 4, a streaming speech recognition model covering 40 language locales from a single 600-million-parameter checkpoint. The Cache-Aware FastConformer-RNNT architecture processes each audio frame exactly once, eliminating the redundant recomputation that conventional buffered-streaming systems apply across overlapping audio windows. Latency is configurable at deployment time between 80 milliseconds and 1.12 seconds with no retraining required. Weights ship on Hugging Face under an open model license and run on Apple Silicon, NVIDIA GPU, or CPU hardware.

NVIDIA reports 17 times the concurrent stream capacity of buffered alternatives on an H100. The Hugging Face release includes a complete fine-tuning recipe covering data preparation, training, evaluation, and deployment; NVIDIA reports a 31% relative word error rate drop on Greek and 32% on Bulgarian from fine-tuning on domain-specific audio. Teams building voice agents for regulated applications, including medical transcription and compliance recording, can fine-tune the model on their own audio without routing a single second of that data to an external server. A NIM (NVIDIA Inference Microservices) deployment package with gRPC streaming support is planned for later this month.

Other speech-related releases from the same week:

• DictaFlow appeared in the Microsoft Store for Windows, offering push-to-talk dictation and screen intelligence for editing text by voice across Windows apps, including remote desktop environments.
• AssemblyAI updated its Universal-3 Pro model, reporting a 19% reduction in errors on multilingual tests and a 30% drop in latency.
• Manus released a meeting notes tool that records, transcribes, and summarizes conversations with automatic speaker identification.
• Google Gboard integrated Rambler, a Gemini-powered feature for real-time speech interpretation and noise suppression on Pixel devices.

The Deployment Choice That Benchmarks Cannot Make

Mordor Intelligence’s 2026 voice recognition market report puts global market value at $22.5 billion this year on a trajectory toward $61.7 billion by 2031, a 22.38% compound annual growth rate. Cloud deployment captured 67.9% of 2025 market revenue, per the same report. The fastest-growing deployment subsegment is on-device and edge voice AI, forecast at 22.96% compound annual growth through 2031. Healthcare providers, at 12.8% of end-user vertical revenue in 2025, are forecast to grow at 23.94% annually, the highest projected rate in the study, with patient-data residency regulations cited as a driver of local-processing adoption across the vertical.

For regulated industries, the decision between cloud and on-device transcription is primarily a compliance determination. A hospital under strict patient-audio regulations cannot route speech data through an external cloud regardless of benchmark rank. European financial institutions subject to data residency requirements generate audio that cannot travel to non-local cloud infrastructure without additional legal safeguards. Local processing sidesteps the routing question entirely. Running locally also means English-only availability for now and a hardware dependency that cloud APIs don’t impose on the buyer.

Microsoft’s cloud API, with broad multilingual coverage and keyword biasing for specialized vocabulary, integrates through Teams and Copilot as a credible backend for customer service, legal document processing, and enterprise communications, so long as Azure’s data residency terms satisfy the compliance requirement. Google’s macOS app addresses a different buyer: one where the device boundary is non-negotiable and waiting on more languages is an acceptable interim cost.

Google has committed to adding more languages to Eloquent but hasn’t said when. Microsoft hasn’t published a general availability date for either new speech model, both currently in limited preview.

The claim arrived with a significant amount of contradicting evidence attached. A Walmart pre-order page for a PS5 version appeared and was removed in the hours before Grubb spoke, the European age-ratings board PEGI had already classified the game for Sony’s console, and retail tracker billbil-kun confirmed a physical PS5 product code with an assigned UPC existed in retail databases at the time of the report. Multiple trails pointed toward PlayStation; one very recent executive call, per Grubb, turned them off.

Grubb Versus the Walmart Listing

Grubb delivered his key claim during the Giant Bomb @ Nite Day 2 livestream that ran hours before Microsoft’s Xbox Games Showcase on June 7. He has covered Xbox closely for years, and the specific phrasing mattered. ‘Just got made’ was not a plan abandoned months earlier but a decision from the days or hours before the Showcase itself.

Gears of War: E-Day was gonna be on PS5. It’s not anymore. That just happened. That was a decision that just got made.

Grubb added on the stream that the company was ‘going backwards on this stuff,’ a skeptical read of the reversal that came alongside the factual claim.

Gears of War: E-Day is a prequel to the long-running franchise, set 14 years before the original game and following younger versions of Marcus Fenix and Dom Santiago through the first days of the Locust emergence on Sera. The Coalition, the series’ developer since the Xbox One era, is building it with People Can Fly, targeting a 2026 release. Xbox gave it a standalone Direct immediately after the main Showcase, confirmed in a March 2026 Xbox Wire post announcing the Xbox Games Showcase 2026 and Gears of War: E-Day Direct, the kind of hour-long slot normally reserved for a franchise the company wants squarely in the spotlight.

Several independent signals had pointed toward a PS5 release before Grubb’s report:

• A Walmart page listed a Standard Edition for PS5 at a $99.99 placeholder price and was later deleted
• PEGI, the European game-age ratings authority, had classified E-Day for Sony’s platform
• Billbil-kun confirmed a physical PS5 retail SKU with a unique product code exists, though the tracker could not confirm the cancellation
• Pre-orders for Xbox Series X|S and PC were expected to open immediately after the Showcase, with no PS5 listing alongside them

Physical retail codes and age classifications require formal publisher submissions that take weeks to process. Their existence in retail systems suggests the PS5 version was in active development for some time before the reported reversal.

The New CEO’s First Platform Call

Asha Sharma’s tenure at Xbox began February 23, 2026, after Microsoft announced her appointment as EVP and CEO of Microsoft Gaming three days earlier. She replaced Phil Spencer, who had led the division for over a decade, and arrived from Microsoft’s CoreAI product division, where she had overseen AI tools for enterprise customers. Before Microsoft, she served as chief operating officer at Instacart and as a vice president at Meta. The selection bypassed Sarah Bond, Spencer’s widely expected successor, and generated intense early scrutiny about whether an executive with no games industry background was equipped to lead the platform.

Her first 100 days produced a run of visible changes. She dropped ‘Microsoft Gaming’ as the division name and returned to Xbox. She cut Game Pass Ultimate from $29.99 to $22.99 per month, a partial walkback of a 50% price hike from October 2025 that had accelerated subscriber loss, as covered in the Game Pass Ultimate price reduction breakdown on this site. She wound down Copilot on Xbox consoles after saying ‘our console players aren’t excited about that.’ She launched Xbox Player Voice, a public feedback portal, in May 2026.

The feedback landed quickly and unevenly. A return to console exclusives topped the board with more than 22,000 votes, ahead of backward compatibility and free online multiplayer. Sharma had already shifted her public language on the question, saying Xbox would ‘reevaluate our approach to exclusivity’ and that successful platforms must have exclusive content and services. She stopped short of committing to halt every existing PS5 plan. Her framing diverged noticeably from the approach Phil Spencer used when the cross-platform push began several years earlier.

The June 7 Showcase was the first under Sharma’s leadership and the most visible test yet of whether her pledges to core Xbox fans would translate into actual game decisions. Per Grubb, E-Day was the franchise where that translation happened.

The Gears Franchise’s Quiet PS5 Debut

The franchise had a PS5 data point before any decision on E-Day’s platforms was finalized. Gears of War: Reloaded, an upgraded version of the original game’s Ultimate Edition, launched on PS5 in August 2025 and became the first Gears title ever released on a non-Xbox console. Its opening player count was announced quickly. Sustained commercial coverage did not follow.

What the experiment showed, set against what was happening across Xbox’s PS5 ports at the same time:

• 1 million players – Gears of War: Reloaded, across Xbox Game Pass and PS5, first three days (August 2025)
• 5 million copies sold – Forza Horizon 5 on PS5 alone, roughly nine months from its April 2025 launch, per Alinea Analytics estimates
• $300 million+ – estimated PS5 revenue from Forza Horizon 5, per Alinea Analytics head of market analysis Rhys Elliott

Gears of War built its audience in the Xbox 360 era on military cover-shooter mechanics and a grim aesthetic that had no real history on PlayStation. Forza Horizon is an open-world racer with worldwide settings and near-universal accessibility; it has never required prior brand loyalty to enjoy. PS5 players could walk into Forza with no backstory at all. Most of them had spent two decades without ever needing a Gears game. The franchise got nowhere near what Forza found on the same platform.

The Coalition’s choice to tell a prequel story gives new players an entry point the series has never offered before. Beginning at the moment the Locust first surfaced, E-Day can function as an introduction to a franchise most PS5 players have never touched. But whether that story design would translate to new PS5 fans was exactly what Reloaded was supposed to test, and the experiment ended with coverage moving on fairly quickly.

Xbox Draws the Exclusivity Line at Consoles

One detail stops this from being a full reversal to the old exclusivity wall. PC is staying in. Gears of War: E-Day will launch on Xbox Series X|S and PC, which means the bet is a console-differentiation play, not a total withdrawal from all platforms outside Xbox hardware. Windows is a Microsoft platform, and Xbox has kept it in the release plan for first-party games throughout the multiplatform era.

With PC staying in, the decision is specifically about what PlayStation gets. The argument, as Grubb’s framing implies Sharma is making it, is that flagship franchises most coded as Xbox identity should not hand their new mainline entries to a competing console while Xbox hardware has nothing uniquely its own.

The 2026 slate adds pressure to that argument. As reported on this site, Fable slipped to February 2027, leaving the fall window thinner than planned. Halo: Campaign Evolved is still confirmed for PS5. Forza Horizon 6 launched May 19 on Xbox and PC with a PS5 port coming later. Call of Duty continues on every platform. E-Day is the single major first-party title sitting in the Xbox-only column for 2026.

Microsoft does not publish console unit sales, and the company has been candid that its gaming segment increasingly runs on Game Pass subscriptions and software. That commercial structure is one reason the multi-platform push started in the first place. A console-exclusive title is a lever that can change the hardware picture, but it only works if the franchise can pull PS5 owners across the platform divide, and Gears has never been tested at a new-release level on that hypothesis.

The $300 Million Argument Against Going Xbox-Only

The counter-argument is sitting in a spreadsheet inside Redmond. Forza Horizon 5 crossed an estimated $300 million in PS5 revenue by January 2026, per Alinea Analytics. Two million copies of the port sold in its first month after the April 2025 PS5 launch. That figure came from a game that had originally shipped for Xbox and PC in November 2021, four years earlier.

The franchise-by-franchise picture from Xbox’s PS5 years:

The pattern cuts along franchise lines. Service games and broadly accessible titles benefit from the widest player pool. E-Day is a single-player military prequel with a limited PS5 track record and no history of building new fans on Sony’s platform. The internal case for keeping it off PlayStation is easier to make when the franchise’s own PS5 data does not offer much of a counter-argument.

Grubb raised the downside plainly on stream, questioning whether the strategy can move hardware at the scale Xbox needs given the platform’s install base disadvantage. Microsoft’s most recent quarterly report showed Xbox hardware revenue down roughly 33 percent year over year. For the call to pay off, E-Day needs to push people into buying an Xbox, something no single first-party game has managed at meaningful scale in recent console cycles. Reloaded did not build the PS5 profile that would make new console purchases likely.

The Gears of War: E-Day Direct follows the Xbox Games Showcase on June 7, with pre-orders expected to open immediately after on Xbox and PC. The platform list at that moment will be the first official word on whether Grubb’s report holds.