> daily_signal(2026_05_06)

A 5-year-old chatbot company just spent $200 billion on computer chips.

PickBits Daily Signal · Wednesday, May 6, 2026

By Mark Pickering · 6 min read · May 6, 2026

// tl;dr

Anthropic just spent more money on computer chips than the war in Afghanistan cost. Anthropic, the company that makes the Claude chatbot, promised Google $200 billion over five years for chips and electricity. They also took over a warehouse in Memphis with 220,000 of the most expensive computer chips on Earth, built by Elon Musk. Two months ago the Pentagon banned Anthropic for refusing to let its AI be used to control weapons. Today they spent more money than most countries have.
The federal government just sided with Elon Musk to take away your right to ask why an AI rejected you. Last year Colorado passed the only US state law that forced AI companies to explain how their software denied your loan, threw out your job application, or flagged your insurance claim. The Department of Justice just filed paperwork to help kill that law. Colorado's own state lawmakers are gutting it from the inside.
Apple just agreed to pay you $25 if your iPhone could not do what they said it could. A federal judge approved a $250 million settlement yesterday. About 37 million iPhone 16 and iPhone 15 Pro buyers qualify. The reason: Apple sold those phones with ads promising a smarter Siri the company knew was two years from working. Apple is paying but not admitting they did anything wrong.
Counter-signal: Jensen says AI is creating jobs. Microsoft just offered 8,750 of its own employees a buyout to leave. NVIDIA's CEO told a Wall Street audience on Monday that AI is “creating an enormous number of jobs.” Tomorrow morning, Microsoft sends letters to 8,750 US workers asking them to take a deal and go. AI teams at Microsoft are explicitly not allowed to take it. Both stories are true. They run on different sides of the same desk.

Five things landed in the last few days that change how AI shows up in regular people's lives. A chatbot company nobody had heard of five years ago is reported to have promised one company $200 billion. The federal government joined a lawsuit to take away the only state law that lets you ask an AI why it rejected you. Apple agreed to write a check to 37 million iPhone buyers because the smart Siri Apple advertised did not actually work. NVIDIA put half a billion dollars down to start making AI cables in America after years of buying them from Asia. And Anthropic shipped a free AI assistant that does most of what a junior banker does. The companies building AI are now spending more money than countries do, and the rules that were supposed to protect normal people from AI decisions are being deleted faster than they were written.

The AI companies are spending more than countries. The rules to protect you from AI are being deleted before they took effect. Same memo.

1. A 5-year-old chatbot company just spent $200 billion on computer chips.

The biggest financial commitment in the history of computing landed in the last 48 hours. The lab that signed it was banned by the Pentagon two months ago. Anthropic makes Claude, a chatbot like ChatGPT. On Tuesday, The Information reported that Anthropic has reportedly committed to paying Google $200 billion over the next five years for computer chips and electricity. Neither Anthropic nor Google has put out a press release confirming the number. This morning, Anthropic's head of product Ami Vora announced on stage at the company's developer conference in San Francisco that Anthropic is also taking over the entire Colossus 1 supercluster in Memphis: a warehouse with 220,000 of the most expensive computer chips on Earth and the electricity to run them. Colossus 1 was built by Elon Musk's company xAI and is now run by SpaceX after the two merged earlier this year. SpaceX confirmed the deal hours later.

The structural detail is the timing. Two months ago, Secretary of War Pete Hegseth labeled Anthropic a “supply chain risk” and the Pentagon banned the company from its eight-vendor classified-network AI contract list. The reason: Anthropic refused to let its AI be used to control weapons. Everyone assumed the ban would cripple Anthropic. The last 48 hours say it did not. Through Google, Elon Musk's SpaceX, and a chip-maker called Broadcom, Anthropic just demonstrated it can buy more computer power than most national governments command, even though the federal government refuses to do business with it.

CNBC coverage of Anthropic Colossus 1 SpaceX compute partnership — cnbc.com · May 6, 2026

Why this matters: When normal people argue about AI, they argue about whether ChatGPT will take their job or write their kid's homework. The actual story is happening way higher up. Companies that did not exist five years ago are placing the biggest financial bets in the history of technology. $200 billion is roughly the entire annual budget of the country of Egypt. One private company, no factories, no products you can hold, just promised to spend that on computer chips. Anthropic was supposed to be the small ethical AI company. They just demonstrated they can write checks bigger than the US Navy's annual budget. The AI companies are not underdogs anymore. They are the biggest spenders on Earth right now, and most of us have not noticed yet. If you are trying to figure out what these companies actually are now and how their decisions land in your industry, that is the pickbits.ai/consulting conversation we are having this week.

https://www.cnbc.com/2026/05/06/anthropic-spacex-data-center-capacity.html

2. The federal government just sided with Elon Musk to take away your right to ask why an AI rejected you.

The only US state law that gave you a right to an answer is being killed from two directions at once. If a computer denies your loan, throws out your job application, or flags your insurance claim, you usually have no idea why. Last year, Colorado passed the only US state law that fixed that. Their 2024 Algorithmic Discrimination Act (SB 205) forced the company running the AI to actually explain how it made the decision in plain English. On April 24, the Department of Justice filed paperwork to help Elon Musk's company xAI kill that law in federal court. The DOJ called the law “unconstitutionally vague” and said it “jeopardizes the country's position as global AI leader.”

While that lawsuit moves through federal court, Colorado's own state lawmakers are gutting the same law from the inside. A new bill called SB 189 drops the explanation requirements entirely and pushes the start date back six months, from June 30, 2026 to January 1, 2027. Same direction as the federal lawsuit, faster path. Governor Jared Polis has been quietly trying to weaken the law since the day it passed. The country had one law on the books that gave you a right to an answer when an AI denied you. Both the federal government and the state are trying to delete it before it ever took effect.

DOJ press release on intervention in xAI Colorado AI Act lawsuit — justice.gov · April 24, 2026

Why this matters: This one is about you, not Silicon Valley. AI is already deciding who gets a mortgage, who gets hired, who gets a payout on an insurance claim, and who gets flagged as risky at the bank. Colorado's law was the only place in America where you had a legal right to make the company explain the decision in plain English. The federal government just tried to take that away. The stated reason was that explaining decisions is “unconstitutionally vague.” The actual reason is that AI companies do not want to write that explanation, and the people running the federal government right now agree with them. Watch Connecticut next. They have a similar bill on the governor's desk right now. If they sign it, the country still has a state-level option. If they do not, you can be denied by an AI and have no path to an answer.

https://www.justice.gov/opa/pr/justice-department-intervenes-xai-lawsuit-challenging-colorados-algorithmic-discrimination

3. Apple just agreed to pay you $25 because they sold you a phone that could not do what they said it could.

The first federal-court settlement that puts a price tag on shipping the AI ad before shipping the AI feature. If you bought a new iPhone between June 2024 and March 2025 — that is the iPhone 16 line and the iPhone 15 Pro — Apple owes you money. On May 5, a federal judge in California granted preliminary approval to a $250 million settlement against Apple. About 37 million phones qualify. Each one gets at least $25 back, up to $95 depending on how many people file. The lawsuit's claim was simple. Apple promoted Apple Intelligence and a new smarter Siri at WWDC 2024 and the iPhone 16 launch as ready-now features. The lawsuit said Apple knew internally that those features were eighteen to twenty-four months from working. Apple delayed the smarter Siri publicly in March 2025. The class period closes the next day. Apple is paying the money but is not admitting they did anything wrong.

The cheap headline number is the story. $250 million sounds like a lot of money. Spread across 37 million phones, it works out to about seven dollars a phone. Apple makes billions selling iPhones. The math tells every product team in Silicon Valley what shipping the press release before shipping the feature actually costs in court. Microsoft is doing the same thing right now with a feature called Recall. Google is doing it with their AI search answers. Salesforce is doing it. The next time you see a TV ad for a new phone or laptop bragging about an AI feature, ask yourself if the feature actually exists yet. Apple just put a price tag on the answer.

AP wire coverage of Apple $250M iPhone AI features class settlement — wral.com · May 5, 2026

Why this matters: Two things, one for you, one for the industry. For you: if you bought one of those iPhones, you can file for the money. The court will set up a website to make a claim. The first round will pay $25 a phone. If not enough people file, the rest goes up to $95 per qualifying phone. That is real cash that is yours. For the industry: every consumer-facing AI product is now operating under a known settlement-cost number. Seven dollars a phone is comfortably under what most companies spend on AI marketing per phone. The lesson Silicon Valley learned this week is that you can advertise an AI feature that does not exist yet, ship the device anyway, get sued, and pay seven dollars a unit to make it go away. That is cheap. The exposure is real but it is also priceable, and Apple just gave the rest of the industry the price. If your team is selling AI features under marketing pressure when the engineering is not actually ready, that is the pickbits.ai/consulting conversation.

https://www.cbsnews.com/news/apple-iphone-settlement-95-payment-how-to-claim/

PickBits.ai is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber at pickbitsai.substack.com.

4. Anthropic just shipped a free AI assistant that can do most of what a junior banker does.

One week after Anthropic announced a $1.5 billion deal with Wall Street to put their AI inside banks, the actual product showed up. Anthropic released ten ready-to-use AI assistants for finance work yesterday. Things like a tool that builds pitch decks. A tool that listens to a company's earnings call and tells you what they said. A tool that balances the books at the end of the month. A tool that flags suspicious transactions for fraud. A tool that writes the paperwork to check whether a customer is laundering money. They run inside Microsoft Word, Excel, and Outlook — the same software every banker already opens at 7 a.m.

The bigger move is the data. Moody's database of 600 million companies is wired directly into the assistant. The AI does not have to look anything up. It already knows. One week ago, Anthropic announced a $1.5 billion deal with Blackstone, Goldman Sachs, and Hellman & Friedman to put Anthropic engineers inside Wall Street firms. Today, the product those engineers exist to deliver shipped. The pieces are connecting in real time.

Anthropic announcement of preconfigured finance agents with Moody's database integration — anthropic.com · May 6, 2026

Why this matters: Banks have spent the last two years building their own version of this. They hired engineers. They paid consultants $1,000 an hour to plan it. They spent eighteen months and tens of millions of dollars on a custom version. Anthropic just shipped the finished version as a one-click install. If you work in finance, the people who used to be called the “AI strategy team” at your bank are about to have a very awkward Monday meeting. For everyone else, this is the canary in the coal mine for white-collar AI replacement. The first batch of jobs that ChatGPT does not really do well, but ten purpose-built AI assistants plugged into your company's systems absolutely do. Pitch decks. Earnings call summaries. Bookkeeping. Fraud screening. Month-end close. Every one of those used to be a junior-banker job. If you have to walk into a budget meeting next week and explain whether to keep building or just install the version Anthropic shipped, that is the pickbits.ai/consulting conversation.

https://www.anthropic.com/news/finance-agents

5. NVIDIA just put $500 million down to start making AI cables in America.

Almost every cable inside an AI data center comes from Asia right now. NVIDIA just made the first move to change that. The chips that run AI live inside giant computer warehouses called data centers. Inside those warehouses, every chip is wired to every other chip with skinny glass cables. The cables are how the AI talks to itself. Right now, ninety percent of those cables are made in Asia. On May 6, NVIDIA — the company that makes the AI chips — announced a long-term partnership with Corning with $500 million committed: fifteen million Corning shares at $180 each, plus three million pre-funded shares. Three new factories are coming. Two in North Carolina and one in Texas. Three thousand new manufacturing jobs. Corning's stock jumped 14 to 20 percent the day they announced it.

This is the AI version of the chip factories Taiwan Semiconductor is building in Arizona. Same idea. Bring the part you cannot afford to lose back home. The chips get all the headlines because Jensen Huang is on TV talking about them. The cables are the part nobody talks about. NVIDIA needs both. Right now, if a hurricane hits Taiwan, or a trade fight breaks out, or a ship gets stuck in a canal, the cables stop showing up — and every new AI data center in America stops dead. Three new plants in two states is not a press release. It is a 24 to 36-month build, with a ten-year supply implication.

NVIDIA newsroom announcement of Corning partnership for US AI manufacturing — nvidianews.nvidia.com · May 6, 2026

Why this matters: This is the part of the AI story almost nobody talks about. The chips get all the attention because they are expensive and the chip CEO is a TV celebrity. But the chips are useless without the cables that connect them. NVIDIA just paid up to make sure the cables can be made here. For people in North Carolina and Texas, that is real manufacturing jobs starting in 2027 — the kind of jobs that come with overtime and benefits. For everyone else, this is the cleanest sign yet that the AI buildout is becoming an actual industrial project. Physical factories. Hard hats. Regional economies. Not just code on a screen. The AI conversation in your living room is mostly about ChatGPT. The AI conversation that is about to land in your county is mostly about who gets to build the warehouses and who gets to staff them.

https://nvidianews.nvidia.com/news/nvidia-and-corning-announce-long-term-partnership-to-strengthen-us-manufacturing-for-ai-infrastructure

Δ Counter-signal: Jensen says AI is creating jobs. Microsoft just offered 8,750 of its own employees a buyout to leave.

Jensen Huang, NVIDIA's CEO, told a Wall Street audience at the Milken Institute Global Conference on Monday May 4 that AI is “creating an enormous number of jobs.” That is the second high-profile AI CEO running the same talking point in roughly a month. Sam Altman ran the first one in mid-April. The framing is buying real things. NVIDIA's own $500 million Corning deal that anchors story #5 above lists 3,000 new manufacturing jobs as a deliverable.

Inside the same week, Microsoft sends letters tomorrow morning to roughly 8,750 of its own US employees offering them a deal to leave the company. The deal is targeted at people who have been at Microsoft the longest and know the most about how the place runs. The teams that build AI products at Microsoft — Copilot, the chatbot, the image generators — are explicitly not allowed to take the deal. They keep their jobs. Everyone else gets a 30-day window to decide whether to take the buyout or stay.

Both stories are true at the same time. Building AI is creating jobs — the NVIDIA-Corning deal is real, and the manufacturing work in North Carolina and Texas is real. Using AI inside an existing big company is killing jobs — Microsoft needs Copilot to actually replace work or its giant AI bets do not pay off, so it is shrinking everywhere except the AI teams. The same companies are running both plays simultaneously. NVIDIA needs Corning to staff three new plants. Microsoft needs Copilot deployment to justify shrinking its corporate headcount. When a CEO tells you AI is creating jobs, the question to ask is “where, doing what, and who at this company is also being shown the door this week.” The answer is usually different at the building site than at the office.

TechCrunch coverage of Jensen Huang Milken comment that AI is creating enormous jobs — techcrunch.com · May 4, 2026

https://techcrunch.com/2026/05/04/as-workers-worry-about-ai-nvidias-jensen-huang-says-ai-is-creating-an-enormous-number-of-jobs/

» What to watch this week

Microsoft buyout letters go out tomorrow. 8,750 US workers get the offer May 7. They have 30 days to decide. The number who say yes will become public at the next earnings call. That is the real signal on what AI is actually doing inside one of the biggest companies in America.
The $200 billion Anthropic-Google number gets a real source or it does not. Right now, only one outlet (The Information) has reported the figure. If Anthropic or Google confirms it on the record — press release, earnings call, regulatory filing — the number is real. If they stay silent for two weeks, the number is probably overstated.
Apple settlement claims website opens. Final court approval and the website where you actually file for your $25 are the next two steps. If you bought an iPhone 16 or iPhone 15 Pro between June 2024 and March 2025, watch for the link.
Colorado governor signs SB 189. If signed, the only US state law that gives you a right to ask why an AI rejected you loses its core protection before it ever took effect.
Connecticut governor signs SB 5. Connecticut's similar bill is on Governor Ned Lamont's desk. If he signs it, the country still has one state-level law that gives you that right. If he does not, that option is gone too.

Tomorrow's signal lands here.