DealBook: Warsh’s market boost
Also, a scoop on a new government fellowship for engineers.
DealBook
October 16, 2026

Good morning. Andrew here. Consider this: The biggest tech companies will spend around $800 billion on A.I. infrastructure this year. The consensus forecast for next year is more than $1 trillion — a larger share of G.D.P., according to Goldman Sachs, than in any investment cycle since the railroads.

Washington has decided that A.I. is a national priority. But almost nobody is asking a simple question: Why are we paying so much for it, and could it be done for less? Nvidia’s gross margin last quarter was 75 percent. Micron, which sells memory chips — a commodity it was losing money on three years ago — just reported a margin of nearly 87 percent.

I’m not accusing anyone of gouging. Shortages set prices, and this is a shortage. But a 75 percent gross margin is usually a flashing sign that summons competitors, who compete the margin away. In the A.I. build-out, that light has been flashing since 2023.

The rebuttal, of course, is that the market is already fixing this. We’re watching the rise of AMD, custom silicon from Amazon and Chinese open-weight models. But that will all take time to affect costs. The trillion dollars is being spent now, some of it with tax credits.

We argue endlessly about what a drugmaker should charge. If A.I. is going to be ubiquitous, and if it’s a national priority, then determining the right amount of competition is a policy question. And we should be talking a lot more about that. (Was this newsletter forwarded to you? Sign up here.)

Kevin Warsh, right, in a dark suit and tie, sitting in a white chair against a blue backdrop. Kristalina Georgieva, left, sitting in a white chair with her back to the camera.
Kevin Warsh’s nonanswer on interest rates is giving stock investors some comfort. Lillian Suwanrumpha/Agence France-Presse — Getty Images

The Fed’s calming effect

The S&P 500 appears set to open in the green today, thanks largely to one man: Kevin Warsh, the Fed chair.

Warsh’s comments overnight at the annual meeting of the I.M.F. persuaded many investors that the U.S. central bank wouldn’t raise borrowing costs at its rate-setting meeting this month.

But the big question is when the Fed and others will be forced by persistent inflation to raise rates, and by how much.

“I don’t prejudge meetings,” he said at the I.M.F. event in Bangkok, referring to the Fed’s gathering in two weeks. Reiterating his aversion to forward-looking guidance, he added, “We’re not in that business anymore.”

That was enough to allay investor fears that any unexpectedly hot consumer and producer inflation data would force the Fed to raise borrowing costs. Traders put the odds of rates remaining steady at just under 83 percent, up sharply from yesterday.

Investors are on edge about elevated interest rates, especially now that they’ve become spooked by how much debt artificial intelligence companies have amassed.

Some investors have responded by rotating out of tech stocks and into other sectors like health care and consumer products.

But Warsh and other policymakers warn that inflation will take its toll:

  • At the I.M.F. meeting, Warsh called rising energy prices and vertiginous A.I. spending levels a “shock” to the economy. Perhaps more important, he emphasized that he was committed to “price stability.”
  • He also noted that other central banks around the world were raising rates to deal with inflation.
  • Kristalina Georgieva, the I.M.F.’s managing director, who spoke with Warsh onstage at the event, said that “central banks should be laser-focused on price stability.” She added, “Rate hikes today are better than falling behind the curve.”
  • Beth Hammack, the president of the Cleveland Fed and a voting member of the Fed’s rate-setting committee, said yesterday that inflation expectations might require “more” from the Fed.

And fears about A.I. companies haven’t dissipated. Shares in CoreWeave look set to open lower, after making a comeback yesterday, thanks largely to supportive comments by Jensen Huang, Nvidia’s C.E.O.

But investors continue to fret about the financial health of A.I. companies that aren’t hyperscalers like Microsoft or Alphabet. So-called neoclouds have drawn fresh scrutiny given their high debt levels, rising interest rates and questions about whether A.I. spending will continue at its current pace.

HERE’S WHAT’S HAPPENING

Trade negotiations between the U.S. and Mexico reportedly stall. Commerce Secretary Howard Lutnick has objected to reducing tariffs on imports of Mexican automobiles and metals, mirroring objections that derailed trade talks with Canada, according to The Wall Street Journal, citing unidentified sources. Elsewhere, Taiwan plans to tighten export controls on artificial intelligence chip sales to cooperate with a U.S. push to clamp down on the smuggling of advanced processors to China.

Some investors are said to lower their expectations for Anthropic’s I.P.O. Several investors who attended presentations that the A.I. giant held this week are now discussing valuing the company at around $1.5 trillion, down from earlier expectations of $2 trillion, Reuters reports, citing unidentified sources. The reason, Reuters adds: an apparent slowdown in Anthropic’s quarterly revenue growth.

U.S. mortgage rates remained at three-year highs. The average rate for the 30-year, fixed-rate mortgage, the most popular home loan in the United States, held steady for an eighth straight week at 7.4 percent, the mortgage finance giant Freddie Mac said yesterday. It remains at the highest level since 2023. Rising rates have caused the housing market to slow. Existing-home sales fell 2.2 percent in September from the year before, the National Association of Realtors reported on Tuesday.

A stock chart showing 30-year fixed-rate mortgage rates.
Source: Freddie Mac. The New York Times

Scoop: Cognition co-sponsors a federal digital fellowship

Artificial intelligence is rapidly transforming private businesses. But the federal government needs help modernizing, too — and a new fellowship aims to help.

The A.I. coding start-up Cognition and the National Design Studio, an organization founded last year by executive order, announced the American Forge Fellowship for engineers to tackle some of the government’s biggest digital challenges today, Sri Muppidi is first to report.

More on the fellowship:

  • An initial six engineers — at any stage in their careers — will be selected for a six-month residency at the National Design Studio in Washington.
  • Cognition will provide an undisclosed amount of financial support to each fellow

“It’s just harder sometimes for the government to recruit in engineering versus in Silicon Valley,” Russell Kaplan, a founder of Cognition and its president, told DealBook. “We were trying to brainstorm, what are ways we could help attract really strong engineers to come help the country?”

The fellows will work on projects for America.gov, an A.I.-powered website that helps Americans access federal information and services — including for passport applications, enrollment in Medicare and more — from across 29,000 government websites.

America.gov, which was built by the National Design Studio for the General Services Administration, provides A.I. generated responses to users’ questions via a chatbot interface.

“We have our plates full for the next couple of years really making the super-app for the government, one place where you can get anything done,” Joe Gebbia, the U.S. chief design officer and a founder of Airbnb, told DealBook. “It’s going to take the work of many fellows to pull this off.”

Since America.gov’s debut at the end of last month, “agencies are knocking at the door to integrate,” Gebbia added.

Cognition already works with the federal government. The company has worked with the U.S. Navy and NASA, Kaplan told DealBook. He added that the fellowship could further strengthen the start-up’s ties with Washington.

VISUALIZATION OF THE DAY

A series of spheres representing the number of calculations involved in training different A.I. models
The New York Times

Training artificial intelligence models is an exercise in increasingly mind-boggling amounts of calculations. So Aatish Bhatia of The Times’s Upshot helped visualize the scale of the math.

For reference, OpenAI’s Astra model, released this year, required nearly an octillion additions and multiplications. (That’s a thousand trillion trillion, or a thousand times the number of stars in the known universe.)

That has a price. Training Grok 4, a model form xAI, required about half an octillion calculations, according to the nonprofit Epoch AI. That cost about $490 million in hardware and electricity and released as much carbon dioxide as the annual carbon footprint of 10,000 Americans.

Big Law’s hunt to cash in on A.I.

The giant law firm Freshfields had a major “oops” this week, when its global managing partner, Alan Mason, accidentally forwarded an email exchange with a top partner, Ethan Klingsberg, to a broad internal distribution list.

Being discussed: a preliminary idea to spin out Freshfields’ artificial intelligence division, potentially for a valuation of “billions,” The Financial Times reported.

It’s the latest example in a push among big law firms to cash in on the A.I. boom beyond traditional legal practice, Niko Gallogly reports.

The context: As A.I. threatens to upend the legal world’s business model of billable hours, firms have been racing to take advantage of new tools. Kirkland & Ellis, for instance, has committed to spending $500 million to build its own A.I. platform.

Others have partnered directly with frontier labs. (In April, Freshfields entered an agreement with Anthropic to build an in-house tool that could later be sold to other firms.)

Many big firms have struck deals with start-ups like Harvey and Legora, which provide A.I.-powered legal services. Harvey is now valued at $15.5 billion, and Legora at $5.5 billion.

Law firms want in on those valuations, Kimball Parker, the C.E.O. of the A.I. legal services company SixFifty, told DealBook. “The multiples that you can get on a software business or an A.I. business are dramatically larger than you can get on a service business,” he said.

The catch: Most states require law firms to be owned by lawyers. To bring in outside investment and unlock higher valuations, law firms are spinning out businesses and finding other ways to get creative:

  • SixFifty was a subsidiary of the Silicon Valley law firm Wilson Sonsini that was sold last year to the payroll company Paychex for an estimated $70 million to $85 million.
  • The venture-backed start-up Eudia took advantage of looser law firm ownership laws in Arizona to open an “A.I.-augmented” practice.
  • Some major law firms — reportedly including Paul Weiss, Quinn Emanuel and Proskauer — have explored selling stakes to private equity firms, a move that could give practices money to invest in A.I.

We hope you’ve enjoyed this newsletter, which is made possible through subscriber support. Subscribe to The New York Times.

THE SPEED READ

Deals

  • The British government is said to have courted UBS to relocate to London if the Swiss bank decides to leave Zurich because of strict new capital requirements. (FT)
  • Companies owned by private equity firms are increasingly taking on debt to fund dividend payments to their owners, as investment firms race to return cash to their limited partners. (CNBC)

Politics, policy and regulation

  • “Most of Ken Paxton’s Money Can Be Traced to One Man: Elon Musk” (NYT)
  • The F.A.A. is said to be seeking a stake in Air Space Intelligence, an air traffic control artificial intelligence start-up whose investors include Andreessen Horowitz. (Politico)

Best of the rest

  • Elon Musk unveiled Tesla’s long-awaited new Roadster model — equipped with air thrusters that let it hover above the ground — in a test of consumer appetite for expensive electric sports cars. (NYT)
  • “The Great Louvre Robbery Was Just the Start. It’s Been the Year of the Heist.” (WSJ)

Thanks for reading! We’ll see you tomorrow.

We’d like your feedback. Please email thoughts and suggestions to dealbook@nytimes.com.

Follow DealBook on Instagram: @nytdealbook