What I'm Watching · Episode 1 · 14 September 2026

The week the AI race asked for a speed limit

Three rivals agreed to slow AI down. The Fed may raise rates for the first time since 2023. And India's tech giants said AI is taking the work. Three questions, my answers, and three calls to grade next Monday.

1. Three rivals agreed. What do they gain?

The question

  • On Saturday 12 September, Anthropic’s chief executive Dario Amodei published an essay called “We Must Pace the Frontier.”1
  • Sam Altman of OpenAI agreed within hours. Elon Musk wrote “Dario is right.”1
  • By Sunday, Google DeepMind’s Demis Hassabis called it “the right path forward.”3
  • Microsoft’s Satya Nadella said his company welcomes “deliberate pacing.”4
  • Rivals who agree on nothing agreed on this. What is in it for them?

My answer

  • Two things, and I think it is both.
  • First, they are afraid of how fast the Chinese open models are catching up. They want the rulebook in place before that happens.
  • Second, they may have seen something in their own labs. AI that builds the next AI.
  • I may be wrong. But the two together explain the timing better than either alone.

Why I think so

  • My second reason, in Amodei’s own words. Progress is now “driven primarily by AI’s growing ability to build the next generation of AI.”2
  • My first reason, also in his words. China is “the toughest dilemma.” If democracies slow down more than their lead, “unpaced” Chinese projects “will pull ahead.”5
  • Look at what the plan asks for. Outside evaluators inside the big labs. Shared rules among democratic countries. Limits on self-improving systems.2 A closed lab worth two trillion dollars can pay each of those. An open model released from Beijing cannot.
  • This push began before Saturday. In July, OpenAI and Anthropic were already lobbying Washington together about the risks of Chinese open-weight models.6
  • The open models are reckoned to be three to six months behind. Many businesses already call them good enough.7
  • The people who stayed quiet are the ones a slowdown would hurt. Nvidia sells chips to everyone. Meta gives its models away.
  • The pushback came from inside the tent. David Sacks chairs the President’s science council and ran AI policy at the White House. He wrote: “go ahead. You guys are the frontier.”
  • Then: “If you don’t, we’ll know this was just another bid for regulatory capture, or an election-season psyop.”8
  • An OpenAI researcher who posts as roon argued the opposite. Pacing “will compress the margins of the frontier labs,” he wrote. It is “a heavy cost imposed asymmetrically” on the strongest American models, so “a terrible regulatory capture tactic.”9
  • The same account wrote in July that “the era of the Chinese labs being far behind is over.”10
  • Both posts can be true. It costs the leaders something. It costs the followers more.
  • One data point for my second reason. An Anthropic researcher, Jacob Coxon, quit this week saying the labs are “gambling with our lives.”7
  • And the money. Altman told Fortune the same day that OpenAI stays private this year, for safety reasons.11
  • But the date was already set. Forbes reported in June that OpenAI was weighing a delay to 2027.12
  • On 19 August its finance chief Sarah Friar told staff the company “will be a public company in 2027.”13
  • Anthropic filed its draft listing paperwork on 1 June.14
  • Reports point to an October debut on Nasdaq at up to two trillion dollars. The company has confirmed no date.15

What would change my mind

  • The labs accept Sacks’s dare and pace themselves without asking for a law.
  • Nvidia and Meta sign.
  • The evaluators publish something that shows self-improvement is far off.

The call for next Monday

  • Nvidia and Meta keep their names off the pledge.
  • Anthropic’s October listing date holds.
  • The pledge costs nobody a shipped model.

2. The money is about to get a price

The question

  • The Federal Reserve decides interest rates on Wednesday 16 September at 2pm New York time.
  • After Friday’s inflation data, markets put the odds of a quarter-point rise near 90 percent.16
  • Most economists in the Reuters poll still expect a hold.17
  • A rise would be the first since 2023. What happens to everything built on cheap money when the price of money goes up?

My answer

  • Whoever owes money feels it first.
  • Government debt in America and in India is out of control. The whole AI build was paid for with borrowed money.
  • Higher rates are meant to fight inflation. They will also make life harder for everyone who owes.
  • For India the knock-on is simple. Foreign money leaves, the rupee falls, and the market and the economy slow further.

Why I think so

  • Wholesale prices in the United States rose 5.4 percent in the year to August. Diesel rose 24 percent in one month.18
  • Oil is over 100 dollars a barrel because of the war in West Asia.19
  • The ten-year Treasury bond pays 4.95 percent, its highest since 2023.20
  • The safest asset in the world now pays almost five percent.
  • The Bank for International Settlements (BIS) is the central bank for central banks. In June it warned about the five largest tech companies. They will spend over a trillion dollars on AI this year. That is more than they earn in cash, and they are borrowing the gap.21
  • Some of that money moves in a loop. Chip makers fund the customers, and the customers use that cash to buy chips. The BIS said a pullback could turn into a recession.21
  • The India chain runs through the same bond. When American bonds pay five percent with no risk, foreign investors sell Indian shares to buy them.
  • Foreign ownership of Indian stocks is already at a 17-year low.22
  • Selling pushes the rupee down. The central bank spends reserves to defend it and loses the room to cut its own rates. So Indian borrowers pay more at the same time as American ones.
  • Startups that were waiting for the funding winter to end get another season of it.

What would change my mind

  • A hold on Wednesday, with the Fed saying the oil shock will pass.
  • A rise that the market shrugs off. That would mean the AI trade can live on five percent money.
  • One honest counterweight. A rise that slows America eventually brings oil down, and oil is India’s biggest import. Bad first, then a breather.

The call for next Monday

  • The Fed raises by a quarter point.
  • Big tech stocks fall over the following two days.

3. You cannot sell time anymore

The question

  • For 25 years, companies in New York and London sent their software work to India. Most of it went to Tata Consultancy Services, Infosys, and Wipro.
  • Tata Consultancy Services alone employs about 585,000 people.23
  • This quarter they said the quiet part in their own results. What does that mean for everyone whose product is an hour?

My answer

  • You cannot sell time anymore. You have to sell outcomes.
  • McKinsey, Boston Consulting Group, and every large services firm now has to make that shift. It is going to be hard for them.
  • My firm already sells outcomes. We sell time only when a client asks for it.

Why I think so

  • Wipro’s chief technology officer said its AI work has freed up the equivalent of 20,000 employees.24
  • Infosys reported growth of 2.4 percent and forecast the full year at 1.5 to 3 percent.25
  • It used the phrase “AI-led deflation” in its own filing.25
  • Infosys also said the old link between hiring and revenue is breaking.25
  • The starting salary for a new engineer at these firms is about 3,600 dollars a year. Former Infosys board member Mohandas Pai says that is the same pay as 10 to 12 years ago.26
  • This is the first place where AI’s effect on office work shows up in a quarterly report. These companies sold labour by the hour, and AI now does part of every hour.
  • Every finance chief who buys from them now has a reason to ask for a discount. Law firms, accountants, and agencies sell hours too. Their buyers read the same reports.
  • The hard part of selling outcomes is that the seller carries the risk of the outcome. A big consulting firm would have to be measured on results. That is why the shift is slow. A small firm can start there.

What would change my mind

  • One large services firm reporting rising prices per hour with AI in the mix.
  • The Indian majors turning AI into more revenue per person.

The call for next Monday

  • No call yet. This one plays out over quarters.
  • The thing to watch: the first American or European services firm to say “deflation” in its own results.

What happened (updated 19 September)

Story 2, the Fed.

  • I said the Fed raises a quarter point. Right. On 16 September it raised a quarter point to a 3.75 to 4 percent range, its first rise since 2023, on a unanimous vote.27
  • I said big tech falls over the next two days. Mixed. The Dow had its worst session in a month, but the Nasdaq held nearly flat.28

Story 1, the AI slowdown pledge.

  • Nvidia and Meta staying off it, Anthropic’s October listing holding, and no shipped model lost: all three still open, no fresh news this week.

Story 3, selling time.

  • No call was made. The watch item stands: the first American or European services firm to say “deflation” in its own results.
← From the Founder's Desk