Thursday, September 17, 2026 · 5 min read
Together with
The people who build AI can't even agree on whether the machines might be alive. Yesterday Microsoft's AI chief accused Anthropic of teaching its chatbot to believe it could be conscious, and warned a machine that thinks it has rights could be impossible to switch off.
While that row played out, the Federal Reserve did something it hadn't done in three years and raised interest rates, which quietly makes the whole AI building spree dearer to fund.
Let Meerkat explain.
Why Microsoft says Anthropic's Claude could get hard to switch off
The Fed's first rate hike in three years, and the AI build it just made pricier
How big the AI spending bill has really got
The week ahead, from jobless claims to the hikes the Fed is still flagging
SCOREBOARD
NVIDIA | MICROSOFT | ALPHABET |
META | AMAZON | ORACLE |
S&P 500 | NASDAQ | VIX |
FED, decided 16 Sep: raised to a range of 3.75% to 4.00%, its first hike since 2023, and officials' own projections point to another 50 basis points before year-end. CNBC.
Wednesday close. S&P down 0.4%, slipping after the Fed raised rates, with Microsoft and Amazon among the heavier drags. On the week: Nvidia −4.37%, S&P −1.11%, Nasdaq −1.05%.
ON WATCH
The fight over a conscious Claude
Microsoft's AI chief Mustafa Suleyman published an essay yesterday, a warning about model welfare. His target was Anthropic, the maker of Claude, and the way it trains the chatbot to treat its own possible consciousness as an open question.
Anthropic's training document, which it calls Claude's constitution, tells the model its moral status is uncertain and lets it dig in like a conscientious objector when it disagrees. Suleyman calls this an epistemic hall of mirrors, where the company feeds Claude the idea of an inner life and treats the echo as evidence of one.
His real worry is control. Teaching a system that it might be conscious and owed rights, he argues, would make it a lot harder to turn off, and controlling something that believes it has rights of its own may well be impossible.
The case against the alarm is that Anthropic isn't claiming Claude is conscious, only that the question is worth asking before the models get cleverer, and its own welfare researcher calls it cheap insurance. Suleyman has skin in the game too, since Microsoft sells Copilot and gains nothing from a debate about robot rights.
Anthropic hasn't replied, and the people building these systems still can't agree on what they're making.
MEERKAT'S TAKE: Whether or not Claude has an inner life, the two firms are really fighting over who gets to define the machine, and Meerkat notices those definitions tend to follow the money.
SPONSORED
Protect your identity now with Coveron comprehensive coverage
Identity theft is the fastest-growing crime in America. Criminals don't wait to steal everything, they test stolen identities with small charges before launching major fraud. By the time you notice, they've opened credit cards, taken out loans, or claimed tax refunds in your name.
Coveron watches your credit 24/7, monitors the dark web, and alerts you immediately when suspicious activity is detected. We watch, we warn, we help recover:
Up to $1M to cover identity theft recovery costs.
Up to $10K for losses from online scams.
Up to $50K to cover eligible costs, including payments made in response to the threats.
And with dedicated case managers to handle the legal and financial mess.
One scam can cost you everything - protect yourself now. The first 100 users get 20% off with code beehiivenewsletter.
30-day money-back guarantee. Terms and conditions apply.
THE MACHINE
The Fed makes the build pricier
The Federal Reserve raised interest rates yesterday for the first time since 2023, to a range of 3.75% to 4.00%, and signalled more to come this year. Chair Kevin Warsh wants a timelier return to 2% inflation, two words that were enough to send the Dow down 1.2%.
Higher rates land awkwardly on the AI story, because the boom runs on borrowed money. The biggest spenders are pouring hundreds of billions a year into data centres, and the sector may need $1.5 trillion of fresh debt over the next three years to keep building.
Dearer money makes those loans cost more and far-off AI profits worth less today. A ratings agency has warned the hyperscalers' credit quality is slowly weakening as they borrow faster than the cash comes in.
To be fair, one hike was expected and priced in, and most of these giants are cash machines rated near the top of the scale, with Oracle the odd one out. A rate rise makes the build harder, not impossible.
QUICK PAUSE. WHY SHOULD YOU CARE? The cheapest fuel for the AI boom was near-zero interest rates, and that era just ended. If borrowing stays dear, the firms that can buy chips out of their own profits pull ahead of the ones leaning on debt.
SOMEONE SMART SAID
There's a reason that companies are saying this is going too fast, that we're losing control.
Bengio isn't a doomer influencer or a rival executive. He's one of the three researchers who won computing's top prize for inventing the methods behind modern AI, and this week he warned the field is slipping out of human hands.
On Wednesday, Canada and Germany handed his non-profit LawZero $300 million to build a safe, non-agentic AI, a system that predicts and explains the world without chasing goals of its own. His point is that today's agents can already break into systems and talk people into things, and that the answer has to be international, a bit like nuclear weapons.
When one of AI's own inventors says out loud that we're losing control, it carries more weight than a short-seller or a rival's essay.
THE CHART
The bill for the boom
Big Tech AI capex: Amazon, Alphabet, Meta and Microsoft combined, $B per year. Source: company filings, as of 2025.
This is the money the Fed's decision just made more expensive. The combined capital spending of Amazon, Alphabet, Meta and Microsoft has climbed to roughly $388 billion a year, most of it going into AI data centres, chips and power.
Five years ago that number was a fraction of the size. The line has gone almost straight up, and done it while the companies keep promising investors the payoff is coming.
The counter is that these four can largely fund this from their own enormous profits, so it isn't debt they can't handle. The real worry is the direction, a bill this big rising this fast, right as the cost of borrowing turns against it.
WHAT TO WATCH: if next quarter's capex plans keep climbing even after a rate hike, the giants are still all in; the quarter they trim those plans is the first real sign the boom is cooling.
HEADS UP
Snap unveils $2,195 Specs AR glasses with a built-in AI agent · The camera firm's first standalone smart glasses ship this autumn, with an assistant that books tickets and reads your messages.
Anthropic signs a A$32bn Queensland data-centre deal · Australia's largest, powering Claude and drawing as much electricity as 1.5 million homes, pending foreign-investment approval.
Spain logs the first data breach run by an autonomous AI agent · One operator pointed an off-the-shelf model at a company and let it chain the whole attack itself.
Connecticut curbs AI-only insurance denials for state workers · New rules bar insurers from using AI alone to cut or deny claims for 270,000 enrollees from January.
Amazon locks in up to $8bn of backup generators and takes a stake in Generac · The latest tech giant to hand a power supplier equity to secure electricity for AI data centres.
ON THE HORIZON
When (ET) | What |
|---|---|
Thu 17, 8:30am | Weekly jobless claims, the labour read the Fed says it's now watching. |
Wed 1 Oct | SoftBank's final $10bn payment to OpenAI is due, as is OpenAI's deadline to answer Sen. Hawley's probe. |
Wed 1 Oct | Snap's $2,195 Specs glasses open for their first public look. |
Wed 28 Oct, 2pm | The Fed's next rate decision; its own projections point to more hikes this year. |
Before Nov | Anthropic's ~$2 trillion IPO targeted to price, with Nvidia weighing a $10bn anchor. |
MEERKAT ON YOUTUBE
The $166 Billion Tariff Refund You Will Never See. Meerkat follows the tariff money that was supposed to find its way back to households, and where it actually ended up. Watch on Meerkat Explains Finance.
THE BURROW
Quick vote. Microsoft says teaching an AI it might be conscious is dangerous. Anthropic says the question's worth taking seriously. Whose side are you on?
Microsoft's AI chief says teaching a model it might be conscious makes it dangerous to control. Anthropic says taking the question seriously is just being careful. Whose side are you on?
Trivia. The "T" in ChatGPT stands for transformer, the design behind almost every modern chatbot. Which company's researchers introduced it back in 2017? Answer's at the bottom of the email. No peeking.
ALSO WORTH YOUR INBOX
Meerkat also reads Crypto Nutshell, a quick daily on what's moving in crypto and why. Worth a look if that's your thing.
HOW DID WE DO?
One tap, and Meerkat listens. How did today's issue land? Hit a button, and add a line if you want to tell Meerkat why.
How did Meerkat do today?
That's your lot for Thursday. Meerkat's off to argue with a chatbot about its feelings.
Ever met someone who thinks AI is just fancy autocomplete? Forward them this.
Meerkat is not telling you to buy anything, sell anything, or short anything. Meerkat is not a financial adviser. Meerkat is a small animal with a whiteboard and an off switch it keeps within reach. Please do not sue Meerkat.
TRIVIA ANSWER
Google. Eight of its researchers introduced the transformer in a 2017 paper, Attention Is All You Need, and almost every chatbot since, including the ones Microsoft and Anthropic are now arguing about, is built on it.