The Fed is the real AI regulator: rates, not bills, picked who builds

Short answer: The most powerful AI policy of the decade was never voted on. When the Federal Reserve took the effective fed funds rate from 0.08% in January 2022 to 4.33% by January 2023, US operating-company IPOs fell from 311 to 38, while five cash-rich giants kept raising capital spending and hit $411.5B in 2025. The price of money picked the AI winners before any safety bill got a hearing.

TL;DR: In 2022 every newcomer that went public in the US raised $6.99B combined, and Microsoft, Alphabet, Amazon, Meta and Oracle spent $157.8B on capex, about 23 times more. Those five generated $602.8B of operating cash in 2025, so they never needed the market the rate hikes closed. The fight over AI safety laws is the show; the rate path is the policy, and builders beat it by needing as little outside money as possible.

Key numbers

What	Number	Source, checked Oct 27 2026
Effective fed funds rate, Jan 2022 vs Jan 2023	0.08% to 4.33%	FRED FEDFUNDS
US operating-company IPOs, 2021 vs 2022	311 to 38 (minus 88%)	Jay Ritter, IPO Statistics
US IPO proceeds, 2021 vs 2022	$119.36B to $6.99B	Jay Ritter, IPO Statistics
Capex of MSFT, GOOGL, AMZN, META, ORCL, 2022	$157.8B	SEC XBRL filings
Same five, capex 2025	$411.5B	SEC XBRL filings
Same five, operating cash flow 2025	$602.8B	SEC XBRL filings
10-year Treasury yield, Oct 23 2026	5.00%	FRED DGS10

The rate hike was an AI law, it just was not labelled one

Washington spent three years arguing about AI bills, model licenses and safety institutes. Meanwhile the one lever that actually decided who could afford to build AI moved without a single AI hearing.

The effective fed funds rate sat at 0.08% in January 2022. By January 2023 it was 4.33%, and it held at 5.33% through mid 2024, per FRED. That is the rate that sets what every startup loan, venture fund return target and IPO valuation is measured against.

The effect on newcomers was instant. Jay Ritter's IPO data, the dataset finance professors use, counts 311 operating-company IPOs in 2021 and 38 in 2022. Proceeds went from $119.36B to $6.99B. The public door for new companies did not narrow, it shut.

Follow the money: the giants never needed the door

Here is the part nobody puts next to the rate chart. We summed the quarterly capex and operating cash flow of Microsoft, Alphabet, Amazon, Meta and Oracle straight from their SEC XBRL filings (the data is in the download below).

Their capex was $130.7B in 2021 and $157.8B in 2022, the year rates jumped. It dipped to $154.2B in 2023, then went to $239.1B in 2024 and $411.5B in 2025. Operating cash flow rose from $289.8B to $602.8B over the same years.

So the rate shock hit the two groups completely differently. A startup that needed outside money met a closed IPO window and repriced venture rounds. A giant that throws off hundreds of billions in cash paid for GPUs out of its own pocket. Same Fed, same rate, opposite outcome.

Who benefits from arguing about safety bills instead

Look at where the attention went. Safety bills, export rules and copyright suits are fights the incumbents can afford to win on any timeline, and we have tracked that pattern in how the AI safety push protects closed models. The rate path never gets that scrutiny because it looks like macroeconomics, not tech policy.

That is the hidden-in-plain-sight part. The mainstream frame says AI is regulated, or not, by Congress. The numbers say AI was regulated in 2022 by the price of money, and the regulation favored whoever already had cash. In 2022 the five giants spent about 23 times what every newly listed US company raised combined.

Our reading, which nobody has proven: the incumbents have no reason to fight this kind of regulation, because it fences off the market for them without a lobbying dollar. A safety law can be challenged in court. A high cost of capital cannot.

The cascade nobody filmed

The rate story has no villain on camera, which is why it works. No company was banned and no model was blocked. Founders simply found that the public market that bought 311 new companies in 2021 bought 38 in 2022, and every private round behind them was priced off that.

That is how a policy decides an industry without anyone noticing: the companies that would have competed just never show up in the data. You cannot count the labs that were not founded, but you can count the IPOs that did not happen, and that number is 273 fewer in 2022 than in 2021.

The steelman, and why it does not change the outcome

The best case for the Fed is simple: it raised rates to kill inflation, not startups, and it has no mandate to pick tech winners. That is true about intent.

It does not change the result. A policy that makes outside capital expensive is, by its arithmetic, a policy that favors whoever does not need outside capital. And the pressure is not over: the 10-year Treasury yield was 5.00% on October 23 2026 per FRED, even with the fed funds rate down to 3.75% in September. Long money is still expensive, and long money is what pays for data centers.

The quiet twist is in the 2025 numbers. IPOs crept back to 90 and $38.97B, and the five giants' capex was still about 10.6 times that. The window reopened, and the gap got wider anyway.

How builders win when money is expensive

If the rate is the gate, the move is to stop standing at the gate. The builders who did best since 2022 needed the least outside money: local and open-weight models instead of renting frontier APIs at scale, rented GPUs at falling prices (see our H100 rental price tracking), and revenue from day one instead of a valuation story.

That is not fatalism, it is a strategy. Every dollar of revenue you earn is a dollar the Fed cannot price you out of. Cheap compute plus open models means a small team can now ship what needed a funded lab in 2021.

And watch the rate like a product spec. When the 10-year moves, your competitors' runway moves with it. We put the bond market next to the AI boom in our 2008 comparison, and more on the opinion desk and the economics hub.

What this means for you

Do one thing today: download the data below, put the fed funds rate next to your own burn and runway, and work out how many months of compute your revenue already pays for. If the answer is zero, that is your first project, not your next round. This is not financial advice.

If you want the filings watched for you, Company Watch ($99/mo) watches the SEC filings, federal contracts and lobbying records of the companies you compete with or sell to and flags every change, so the numbers stay verified as conditions change. We keep fixing it at no extra cost until it does what it promises.

Sources

• Federal Reserve Bank of St. Louis, FRED, Effective Federal Funds Rate (FEDFUNDS), checked Oct 27 2026
• FRED, 10-Year Treasury Constant Maturity (DGS10), checked Oct 27 2026
• Jay R. Ritter, University of Florida, Initial Public Offerings: Updated Statistics, Oct 5 2026 edition, checked Oct 27 2026
• SEC EDGAR XBRL company facts for Microsoft, Alphabet, Amazon, Meta and Oracle (capex and operating cash flow), checked Oct 27 2026