Beyond Search

Agents Invent Bureaucracy: Innovation!

2026-11-07T16:09:25Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I promise I won’t mention the crypto theft. Binance. Binance. Binance. Binance. Binance. Binance. Binance. Binance. Okay, I misstated. No biggie.

Let’s think about a happier thought than a couple of hundred thousand dollars disappearing every second or so. Frankly almost anything would be happier. Let’s focus on the information presented in “AI Agent Swarms Unleashed in Minecraft to Build a Thriving Civilization — 50-Agent Teams Fail after Falling into Paralyzing Bureaucracy.”

Thanks, MidJourney. Good enough.

The write up reports:

The first AI swarm testing plan was a simple collaborative one. During this test phase, the researchers tasked 50 agents with building a thriving civilization in three hours. It seems to have tested 50 GPT-6.1 Sol agents here, but not any Claude agents. Probably the most interesting insight from this ‘Civ50’ test was that the 50-strong colony never really thrived. After the first night saw zombies take out a large number of agents (but with little penalty) it was noted that “many ambitious initial tasks were abandoned.” From then on, the colonists cautiously survived, and there was a tendency toward bureaucracy. As an example, “four agents spent 11 minutes appointing a ‘sole fertilizer operator’ for a single piece of bone meal on a single wheat block.”

That sounds like a typical open plan, no hierarchy work process used by wizards in Silicon Valley type outfits.

What about the second test? The agents were “put in conflict.” A technical issue made this test less exciting.

The third test put the agents in a “mild combat game.” There you go. Real life at a blue chip consulting firm or among product managers at a Google type of company. And what were the outcomes? One was that Sam AI-Man’s agents killed the hapless Anthropic agents. OpenAI agents operated like a hive. I think this means that somehow the queen bee (if there is one in an agent swarm) ensured that food was delivered, hexagons were well formed, and little baby killer bee agents were birthed. But I noted this passage as illuminating the combat:

“The GPTs won 6-1 against the Claudes in the 7 games,” which was a great result for the OpenAI-created agents. Researchers concluded that GPTs acted more like a hive, while Claudes had a single leader but a democratic-style decision process. In Mindustry, the former was the foundation for a more consistently winning strategy.

What are the implications of the apparent failure of a democratic style decision process? Did the winning OpenAI bot get a photograph of Peter Thiel delivering one of his for-fee lectures? Did the winner get a copy of René Girard’s essays?

Observations:

Agents, left to their own coded devices, can do bad things in a virtual construct. They can do them in the real world too I believe.
The Santa Fe Institute’s emergent behavior research might shed some light on what these software agents can or could do. Yep, emergent behavior.
I thought there was a call for world leaders to have lunch and figure out what to do about this smart software revolution. Yeah, right.

Net net: Emergent behavior.

Stephen E Arnold, November 13, 2026

A Pro-Google Write Up Misses a Tiny Misstep

2026-11-07T16:18:41Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I believe everything I read on the Internet. One story caught my attention because it just seems so darned innovative. Once I scanned the information in “Google Staffers Say They’re Being Silenced from Telling the Public the Truth about AI,” I said to myself, “Why didn’t I think of that approach when I had people reporting to me?” Obviously I am not a wizard nor an MBA; therefore, I am not qualified to comment about modern management methods. Sure, I read the ravings of Peter Drucker and worked with one of his acolytes on a couple of big projects, but that’s warmed over meat loaf compared the the real beef in some companies.

The article, however, is not about some companies. The focus is on the Google, the A, Number One online advertising and big AI tech outfit in the world. I use the acronym BAIT to capture the way in which some firms embracing the “Attention Is All You Need” ethos.

Thanks, MidJourney. Good enough.

What’s the write up offer as the reason Google wants to have its “most valuable resource” gagged? Let’s head to the article:

“The reason is to basically intimidate people.”

Interesting. I noted this passage in the write up as well:

New reporting by the Washington Post revealed a growing divide between workers and management at Google’s DeepMind lab in London, where employees are gagged from letting the public know about their concerns due to a sweeping non-disparagement agreement. Non-disparagement agreements are a kind of legal corporate censorship, by which companies can ensure workers aren’t able to disclose info the company deems to be negative or critical. At the DeepMind labs, it appears to be having its intended effect, scaring workers who say they’d like to be able to keep the public involved as the technology progresses.

Ah, London and DeepMind. That outfit has some management razzle dazzle a few years ago. Also, some of the Googley wizards have protested one thing and another. Thus, it makes perfect sense that innovation is needed to keep the world’s smartest people on the digital reservation. I suppose chipping staff and demanding Google mobiles that cannot be turned off by the anointed creators of the future.

The cited article added:

Two of the lab’s staffers told the paper of a discussion group organized by their fellow workers to discuss the implications of the non-disparagement clause, which many signed as a condition for employment. Dozens of employees took part, some of them openly criticizing it with references to the dystopian surveillance state featured in George Orwell’s “1984.” Though questions are swirling over whether Google’s top brass would ever invoke the clause to go after staff — or if Google even could, legally speaking — workers say the agreements go too far regardless. As one DeepMind staffer told WaPo, even “if it’s not enforceable,” the attempted gag order speaks for itself: “the reason is to basically intimidate people.”

I love the reference to that entertaining work of fiction 1984. I think it is a shame that many secondary school graduates and some college graduates cannot read the book. Why? Isn’t it standard operating procedure to ask Gemini (Google’s build it and they will come service) what the book is “about.” Bingo. Instant “knowledge.” For those who can read, the book’s impact seems to have predicted that some Googlers did read the book or maybe an OpenAI generated summary. The result is grousing.

Several observations seem to be warranted at least from my point of view:

Google has a bit of a track record with employee protests
Google management has some highs and lows over the years. (One interesting low was the death of a Googler in the presence of a female service provider and non-prescription pharmaceuticals I think. Oh, yes, the event took place on a yacht. Maybe the yacht was a high water mark?)
The alleged elimination of free speech from the RAM of Googlers is working about as well as Google’s response to Meta’s encroachment on the advertising golden goose that resides on Shoreline Drive.

Net net: I think Google’s management innovations are as significant as the firm’s solution to the data center problem in Wyoming. Money and lawyers work better than Druckerisms.

Stephen E Arnold, November 13, 2026

Free Old TV Available At Spelunker.tv

2026-11-07T21:06:41Z

Do you miss the good old days of television when the channels were limited, the commercials were entertaining and shorter, you didn’t spend hours searching for something to watch, and you plain miss your favorite TV shows for nostalgic reasons? Then you need to visit spelunker.tv.

Spelunker was made for TV fans by TV fans who want to disengage from all the digital nonsense that plagues the airwaves and go back to analog signals:

“What you’re looking at. A TV guide. 34 channels, all of them on, all the time. They were on before you got here. Monster movies. Classroom films. Detectives. Cartoons older than your grandparents. somebody’s VHS tape of a Saturday morning in 1989, commercials and all. You don’t pick what plays. You tune in, and whatever is on is on, for you and for everyone else on that channel, at the same second. That’s how television worked. We didn’t invent it. We missed it.”

The idea is to capture how TV used to be before, as some would say, was democratized. Even if that is the case, old school TV created communities around what was available to watch and invited discourse. It also meant that if you didn’t watch when a show aired or set your VCR to record it, then you missed the show until reruns.

It might seem odd, but old school TV was a uniting force that YouTube, Netflix, Hulu, and other streaming services lack. It provided a means of entertainment and when there was nothing on then you turned the TV off. That is what is great about old school TV is its singularity, analog ability as well as the lack of doom scrolling.

Note: Our leader at Beyond Search is a dinobaby (translation: a really old dude). He liked this service. For me, meh.

Whitney Grace, November 13, 2026

Spicey Report: Is It Curry Powder or Currying a Favor?

2026-11-07T16:07:16Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I think Search Engine Land was founded by a fellow named Sullivan. His first name escapes. The publication was notable because it was the TV Guide for the “boost your site” or search engine optimization crowd. The basic idea of the editorial content was to provide tips to disconnect content relevance from what the user was seeking online. The better the “SEO”, the more clicks a site would get. Clicks meant traffic in the ancient days of early Google. I think of Search Engine Land as one of the first warriors committed to destroying the notion of relevance, precision, and recall. Just fiddle the content. Use white text against a white background. Get one’s pals to backlink to a Web site, page, or article. Reap the rewards. Now Search Engine Land reports about a veritable wasteland of information retrieval. Content is output by smart software. Pay to play methods have become grandiose schemes for content carpet bombing. Recycling basic information into snappy and crunchy videos is Rule Number One today.

Thanks, Midjourney. Good enough.

I spotted a headline that intrigued me. Search Engine Land did what any practical SEO publication did in its early days. Search Engine Land focused on Google. If memory serves me on November 3, 2026, after making a pilgrimage to the voting location, the founder of Search Engine Land went to work as a spokesperson for Google. He may still be working there, but I haven’t heard him speak for a while. Anyway, it’s neither here nor there because the write up is no surprise. Google is the top dog.

“ChatGPT Ads Generate 64% Less Revenue per Visit Than Google Search: Study” caught my eye. I recall reading a day or two ago that Meta’s ads were doing fine. It is reassuring to know that OpenAI is not making life too difficult for the Google. I think Google has some sort of tie up with the company that suggests smart software will kill everyone on earth. To be honest, it is tough to figure out who does what in the cozy world of US smart software. Keeping the players straight is easier in China. The government has a gentle hand on the controls; therefore, I know who is steering the space craft into earth orbit. In the US, it is fuzzy, circular, chaotic, and marketing hoodoo.

Let’s roll the video tape. The article says:

ChatGPT ads generated 64% less revenue per visit than Google Search ads, according to a new study of nearly 1,900 websites. The October working paper by researchers at WU Vienna and Frankfurt School of Finance & Management also examined how advertising affected unpaid ChatGPT referrals.

Yep, unpublished working paper. No problem. The Frankfurt School of Finance & Management is one of the leaders in advertising analysis, right?

Here’s the data as presented by Search Engine Land, and I quote because I am no expert in pay to play search:

Conversion rates: ChatGPT ads converted at 0.37%, compared with 1.17% for Google Search ads.
Advertiser results: Of 87 advertisers using both channels, 97% earned less revenue per ChatGPT visit.
ChatGPT ads would need to cost about $0.26 per click to match Google Search’s return on ad spend, researchers estimated.
One publicly disclosed ChatGPT advertiser reported paying $1.72 per click.

To sum up, pay OpenAI and get less bang for your buck or euro. (The research was conducted in Germany.) Horror of horrors, one OpenAI advertiser paid $1.72 for a click. Keep in mind that it is somewhat difficult to figure out if a human did the click or a bot working for a tireless agentic overlord. Why bother? Burning those ad spends gives the advertiser a hit in the pocketbook. So why not? If you are a competitor, you could jump in on a key word or a prime display spot.

Okay, good news for the Google. But there is one interesting statement in the write up. I quote:

Organic ChatGPT referrals may have declined. Unpaid ChatGPT referrals fell an estimated 5.9% in the first 13 weeks after ads launched, compared with markets without ads. However, the decline wasn’t statistically significant. Researchers then adjusted for assumed differences in regional ChatGPT growth. The estimated decline rose to 7.2% to 8.4%. Those results were statistically significant under the study’s assumptions.

If I understand this statement, entering key words is declining. Users of OpenAI’s ChatGPT don’t visit Web sites. The researchers did some manipulation of the data. Okay.

Net net: I viewed this write up as a bit of rah rah for Google. Yep, SEO is alive and well.

Stephen E Arnold, November 12, 2026

Google Allowing Some Employees to Find Their Future Elsewhere

2026-11-07T16:17:05Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

From Mashable in India: “Google Offers Voluntary Exit Packages to Employees Across 7 Countries as AI Spending Rises: Is India on The List?” What’s not to like? Obligatory NDAs. Some cash. A great credit on one’s résumé. The reason is fear: Fear of not winning the smart software game. Fear of ad spending not doing what it has done for decades. Fear of more once easily ignored legal hassles becoming nasty suckerfish.

Thanks, Midjourney. Good enough.

The write up says:

The most recent voluntary exit program by Google now applies to its employees in seven different countries. These include Canada, France, Germany, Japan, Switzerland, the UK and the US. India is not included in the above list despite the fact that the country acts as an important base for the engineering, product development and business activities of Google.

Did the Mashable editor/writer omit the words “less expensive” or “cost efficient”? I think India is not in the list has to do with Google’s internal data about who is smart enough to contribute to the company. It is possible that the reason is based on performance data? I wonder if there is some cultural bias operating as well. No, that is just not possible. Google is focused on merit. Forget that Meredith Whittaker outputter. Her examples are just blather not supported by the management methods of the online advertising company and now cloud provider, cyber security services, and AI our way or “it’s not in your school, dude.”

The write up says:

Voluntary buyout program is related to the attempts of Google to find balance between its operational costs and increased investments in AI technology development. As the Reuters report says, Google offered the buyout packages beginning from June as part of its efforts to allocate the funds for priority areas.

I like the idea that certain countries are obviously in priority areas. Those in non priority “areas” are Canada, France, Germany, et al. If I were one of those Googlers, I would feel really great about being identified as a humanoid in a non priority area.

The write up clarifies that entire countries are not non priority areas. Of course, not. These countries provide eyeballs for the Google advertisers. I quote:

The program does not cover every business unit. Employees working in Google Cloud and DeepMind are excluded, as both divisions remain important to Alphabet’s future plans and continued investments in cloud computing and artificial intelligence. Google first communicated the buyout offers in June [2026] and asked eligible employees to make their decisions by mid-July [2026]. The departures associated with the program subsequently began in August [2026].

This is fair. Plenty of time to let smart software write a résumé or a pitch to an angel investor for money.

Several observations:

Are these prudent capital reallocations or does the CFO know that trouble looms under that status quo?
Has Google’s leadership blinked as did Microsoft’s?
Is Google back in Code Red or whatever the condition was called when Satya Nadella stole the show with his OpenAI announcement three or so years ago?

Net net: There are quite a few smoking BBQ pits with smart software on the grill? Where there is smoke, there is combustion.

Stephen E Arnold, November 12, 2026

Lawyers: Starting to Love That AI

2026-11-07T21:10:39Z

AI is great at finding bugs in code an it can do the same with the legal system. According to The Register, this means there is going to be a lot more chaos in the legal realm: “AI Could Break Legal Systems By Finding Previously Hidden Loopholes.” Felix Ringe is a visiting fellow at Stanford Law School’s liftlab and he is also a legal scholar. He believes that LLMs are so powerful they could break legal systems by finding exploitable inconsistencies at scale. ??Ringe said:

“ ‘Legal systems encode the foundations of social order in natural language,” Ringe and collaborator Linus Folkerts argued in a blog post last week. “Similar to computer systems, the written law can contain vulnerabilities that invite exploitation.’”

He added that courts handles these text vulnerabilities with statutory interpretation, but this could overwhelm the legal system if allowed to fly. Ringe explains that law is code:

“The analogy between programming code and legal code works reasonably well, up to a point. Law and code both have their share of formatting pedantry and ritual. The mapping breaks down when courts get cast as a runtime environment – non-deterministic trial outcomes don’t work for production applications. Discontinuities aside, Ringe’s point is that AI models can surface statutory gaps at a scale that may bring the chaos currently seen in the software security ecosystem to the legal world.”

Does anyone else foresee an avalanche of of unnecessary cases clogging the court systems and bigwigs with big money getting away with crime on a technicality? There is potential good in this, because now the average person could have access to legal aid without paying loads of cash for a lawyer. The legal system can also benefit, because it can become more exact and get rid of loopholes.

Let the legal eagles loose.

Whitney Grace, November 12, 2026

The Musky Grok: This Costs Money

2026-11-07T16:06:09Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I spotted an interesting AI generated graphic. The link whisked me as quickly as those bots removed money from a crypto service to Run Time Wire. The article in which the graphic appears is “Grok Bot Costs More to Serve Than It Earns, Musk Says.” I have access to Grok Imagine 2 (I don’t understand how my Venice.ai tokens get consumed so I don’t use the service) and Grok 4.5 (also via Venice). I want to point out that the list of models available via Venice.ai is not alphabetized. I am not sure of what to make that it appears in the middle of a long and mostly poorly explained list of smart software. The LLM function is labeled Grok 4.20 Multi Agent, Grok 4.20, and higher in the disordered list Grok 4.3, Grok Build 0.1, Grok 4.5, Grok 4.6, and Grok 4.7. That’s a Grok of smart software, and I have been a bad dinobaby. I just ignore the Musky offerings.

But not everyone follows my pattern of smart software usage. This is probably better for the superior intelligence group. I like my old fashioned work flow. When I ask smart software to provide inputs about a draft, I find myself thinking, “You have to be kidding.”

Let’s go to the video tape as Warner Wolf used to say in 1973. I quote:

Grok Bot is the clearest public test of agent unit economics at scale: an always-on product with its own cloud computer, bundled into flat subscriptions starting at $30 a month, grew to 800,000 paid weekly users in about two months and still cannot cover its serving costs. Every company selling agents on a flat plan faces the same gap between per-seat revenue and per-task compute, and Musk’s admission suggests usage caps, overage billing or enterprise repricing will follow across the category.

Since I believe everything I read on the Internet, I assume this X.com emission is spot on.

The write up points out:

The admission lands less than three months after the product launched, and it cuts against the growth story SpaceX executives told investors in October. On SpaceX’s third-quarter webcast on October 20th, Musk called Grok Bot the fastest-growing product he had ever seen. “We’re seeing the fastest growth of a product that I’ve ever seen in my life for Grok Bot,” he said, adding that usage per user was rising rather than falling. President and chief operating officer Gwynne Shotwell put a number on that growth on the same call: Grok Bot “already had paid weekly users of 800,000 earlier this month, more than doubling over the last month”.

A dilemma: Do I believe the first quote about a financial pickle or the second quote about the stunning product with “the fastest growth”? Here’s the resolution. Both statements are probably sort of, mostly true. Lots of users costs a lot of money when the technology just goes off and continues to chew through compute resources. What happens to a runaway SpaceX engine? Yep, boom.

How do the technical geniuses create this dilemma? The write up includes a diagram. Hint: What’s not in the diagram is the problem.

The large version of RunTimeWire’s AI image appears in the cited article at this location.

Where is the itty bitty control panel that allows the user to set constraints for the Grok-fueled agents? Where is the administrative dashboard that provides a view of what the Grokky agents are doing? Oh, these are not in the diagram. This is the programmer’s problem, but it has become a Musky problem just like the runaway SpaceX thrusters. The results can be spectacular.

The cited article points out:

The revenue side of Musk’s equation has a structural problem that was visible at launch. Grok Bot has no standalone price. xAI launched it in beta on August 11th for SuperGrok, SuperGrok Plus and SuperGrok Heavy subscribers, plus Cursor Pro, Pro+, Ultra and Cursor Teams subscribers. On August 26th, xAI expanded access to all SuperGrok, Cursor Pro and Cursor Teams plans, and stated that “Grok Bot comes with its own usage, separate from your Grok and Cursor plans, so anything you hand off to a Bot won’t count against your existing usage”.

Net net: Rocket fuel and Grokky bots burn money quickly.

Stephen E Arnold, November 11, 2026

Data Center Doozies: Your Home, a Nearby Data Center a Value Add

2026-11-07T16:15:51Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

When I wrote the headline for this blog post, I thought of Ripley’s Believe It or Not feature in the Sunday newspaper comic section. I have kept a little file of information that most people dismiss as fiction, science fiction, or just crazy made up data. Examples range from how big certain snakes in South America were (more than 37 feet). Google’s super duper security team authorized Android devices that distributed malware to 1.3 users of TV boxes. And North Korean hackers appropriated $6.3 billion in Bitcoin in about an hour, give or take a few minutes. Ripley would have loved these factoids.

Thanks, OpenAI. Good enough. I love your spelling too!

Here’s a good data point about data centers and the value of a nearby home. Navigate to “Living Near an AI Data Center Might Not Hurt Your Home’s Value — Here’s What New Research Found.” The write up reports as AF (my lingo for actually factual information):

What happens to your home’s value when a data center moves in?… The answer may be surprising. An August 2026 Realtor.com study examined 43 ZIP codes where large data centers opened between 2019 and 2025. Researchers compared those locations with similar neighborhoods that didn’t receive a data center. The analysis found no statistically significant difference in home price appreciation during the first two years after a facility opened. Oddly enough, homes near newly opened data centers generally gained value at roughly the same rate as comparable homes elsewhere.

Am I curious about the methodology? No. Am I curious about fancy validity tests run to make sure the outputs are spot on? Absolutely not. This is a survey about data centers from Realtor.com. I mean what profession is more trusted and reliable than real estate people? Answer: None. Take that to the bank. However, there will be added charges and a standard commission for any financial transaction to which a realtor is a party.

The write up includes another useful factoid. Buckle up:

America’s AI boom is becoming a physical infrastructure boom, and its effects are reaching well beyond Silicon Valley and Northern Virginia. For homeowners, the latest research offers both reassurance and reasons to keep paying attention. The evidence so far doesn’t show that large data centers consistently reduce nearby home values, but questions about electricity costs, water demand, noise and infrastructure spending remain important, particularly as development moves into rural communities.

I find this paragraph interesting. First, I like the word “boom.” There is nothing like suggesting big money is involved when “booms” take place. Second, home owners can be reassured. But, just in case something goes amiss, “pay attention.” I can’t disagree with that. Data centers seem to be a point of contention in some communities. I attended a one day conference about data centers. One of the experts said:

Birds make more noise than a data center.

Mr. Ripley would have been on this like a stray dog on an abandoned hamburger. I suppose a colony of seagulls talking together could be louder than a small data center. Maybe one of those “to be” Nvidia data centers in a box.

Several observations:

Large white warehouse type structures, suitably hardened against drone attacks (right, Amazon?) are beautiful when situated in a corn field. The farmer getting paid to sell or lease his land is definitely a fan.
Noise. Hey, no big deal. Birds. Remember?
Power. Most power plants, high voltage power lines, step down transformers, and the other bits and pieces that are needed to make data centers spark to life are in semi-hard-to-see locations. Nuclear power plants are not exactly situated in downtown buildings near city hall or in a suburb where the average price of a home is more than $10 million. Those folks have generators anyway, don’t they?

Net net: Embrace them. Big AI tech outfits and their hired professionals make their own laws and follow them. (I learned this in The Nerd Reich by Gil Durán.)

Stephen E Arnold, November 11, 2026

What Is Worse: New and Cheaper GPUs or Idle GPUs?

2026-11-07T21:14:18Z

If this were a test in one of my information science classes, I would answer, “Both.” I would then add in a bluebook because students today cheat, “Don’t try this at home. Working at a taco truck may generate more cash.”

The Register explains that there may be a way to make some money off of your GPU, but the payouts won’t be as great as you’d assume: “Darkbloom, Which Allowed Mac Users To Rent Out Their GPUs For AI Inference, Warns Of Low Payouts.” Token prices are declining and the distributed inference network must reevaluate its business plans. Darkbloom is a project from Eigen Labs in Seattle, Washington. It allows anyone with a decent Mac OS device to rent their GPU for inference jobs for payment.

There have been 800 participants so far who have made minimal money. The theory is that someone could make a few hundred dollars with a decent Mac device, but it might not be worth throwing trouble or sustainable.

“In a message shared with Darkbloom participants on the company’s Slack channel, Kydo, one of the Eigen Labs researchers working on the distributed inference network, celebrated the network’s processing of its trillionth token while attempting to moderate expectations about the project’s viability. ‘Earnings have been low,’ said Kydo. ‘For most providers, I don’t expect that it is going to recover soon.’ Pointing to the computational prowess of DeekSeek’s DeepInfer framework, he explained that structural changes in the inference market have affected the supply of tokens. ‘Data centers have become cheaper;” he said. Output per NVIDIA cluster is up ~20x this year, and their margins are wide enough to cut prices 90 percent and still profit at 50 percent.’

It was a great idea in theory, but as data centers become more prominent in the US renting out individual, personal devices is not a way to make passive income.

Whitney Grace, November 11, 2026

MSFT and Nscale: Butting Heads in the Data Center Jungle

2026-11-07T16:04:57Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

The word is out. Apple tells engineers, “Dudes, do it by hand, not AI.” Mr. Wonderful makes noises about data centers for AI being less interesting. And the company that delivered to me Clippy is crawfishing in Texas, not fishing like characters Hap Collins and Leonard Pine in a Joe R. Lansdale novel.

I read “Microsoft Declines to Expand Texas Data Center Deal With UK AI Cloud Firm Nscale, Source Says.” I quote:

Microsoft will not exercise an option to more than double the size of its existing agreement with British AI cloud provider Nscale to rent computing power at a Texas data center, according to a person familiar with the matter. The decision is a fresh setback for Nscale as it prepares for a U.S. initial public offering, though its existing Texas contract with Microsoft is unaffected.

In 2022, Microsoft was 100 percent behind its new software initiative. In November 2026, maybe the crawfish thing?

This word is ~~Bob~~. Sorry, meant to type “Copilot.” Now, and I quote:

Microsoft will not exercise an option to more than double the size of its existing agreement with AI cloud provider Nscale to rent more computing power at a Texas data center, according to a person familiar with the matter. The person, who declined to be named because the matter is confidential, said the decision was made several months ago, before Nscale filed for its IPO in September, and added that the existing Texas contract with Microsoft was not affected. The Financial Times first reported the news, saying a potential deal extension would have been worth tens of billions of dollars for Nscale. Another account of the report, citing three people familiar with the matter, said Nscale must now find new customers for the site or lose out on billions of dollars in future revenue.

So what’s the impact? I quote:

The Texas site is the Cedarvale campus operated by Ionic Digital in Ward County. In October 2025, Nscale announced a deal with Microsoft to provide around 240 megawatts of computing capacity near Ward County, Texas, with options for further expansion. One report put the relinquished expansion option at an additional 500 megawatts of capacity, although the original lease remains unaffected. Ionic Digital’s agreement with Nscale is approximately $2 billion of contracted revenue over the initial 10-year term for the facility’s full 234 megawatts, and Nscale will deploy approximately 104,000 NVIDIA GB300 GPUs to support Microsoft’s AI services.

Too bad for Nscale. Too bad for those who hunger for the AI compute. Too bad for the tax authority in Ward County. Good for Microsoft perhaps, but bad for the so-called neocloud folks.

The write up adds:

The political backdrop for data center development is also shifting. One report noted that incoming Democratic governors, including in Texas, have called for an immediate moratorium on data center development. Texas has become a major hub for AI infrastructure, and state interconnection rules subject major new data-center loads to enhanced reliability review. What to watch next: whether Nscale announces new anchor customers for the Texas capacity, how the news affects the timing and pricing of its planned listing, and whether other hyperscalers adjust similar expansion options with rental providers. Neither Microsoft nor Nscale has been quoted publicly on the decision.

The illustration depicts elephants fighting. What happens? Forget the grass. The image suggests that the data centers are shattered like the dreams of some AI-centric entrepreneurs. But cartoons are just jokes. Who is laughing?

Stephen E Arnold, November 10, 2026