Anthropic's Hidden Money Network Will COLLAPSE Open AI Competition - Bill Gurley Exposes All!

ELI5/TLDR

Bill Gurley — Benchmark partner, Uber’s first money — is now running a policy institute and promoting his book on building careers from curiosity. The headline-bait about Anthropic is a four-sentence aside in a wider conversation: he says Anthropic is lobbying as aggressively as FTX did at its peak, and he reads the AI industry’s doomerism as cover for regulatory capture. The richer threads are everywhere else — why state-vs-state competition could be America’s edge, why China going open-source is partly an answer to twenty years of being called IP thieves, and why the same near-zero-rate decade that made every VC a billionaire also locked in the most risk-seeking capital he’s seen in his career.

The Full Story

The Anthropic line, in context

The thumbnail promises an exposé. The substance, in Gurley’s actual words:

The only company that has lobbied as much as Anthropic early in its life was FTX and SBF. They have people on the ground in every state. They’re the ones pushing for state-by-state regulation… I was worried that the AI people were begging for regulation to pull up the ladder.

That’s it. No spreadsheet, no donor map, no named operatives. The claim is that Anthropic’s policy spend rhymes with the most cynical lobbying push in recent memory (FTX trying to get crypto rules written in its favour before it imploded), and that Dario Amodei’s public anxiety about AI risk is conveniently consistent with wanting a moat dug by Washington. He stops short of calling it corruption — “I don’t think any of them think it’s corruption. I think they’re just playing the game on the field” — which is the most Gurley sentence in the whole conversation.

The “hidden money network” framing is Bilyeu’s. The actual argument is the same one Gurley has been making for years: regulation is the friend of the incumbent, and the incumbent in AI now has the budget and the relationships to write the rules.

Regulatory capture as the through-line

This is what Gurley actually wants to talk about. He won an audience for it at All-In with a forty-minute talk titled “2,851” — the distance in miles between Sand Hill Road and Washington DC, which he intended as a punchline about why Silicon Valley worked. Since giving the talk, he says, the Valley has rooted itself in DC harder than ever. His policy institute is built around three ideas:

Regulation is mostly written by the regulated. George Stigler’s 1958 Nobel work; Gurley’s go-to frame. In finance, healthcare, telecom, the pattern is: incumbents staff lobbying shops with people who recently ran or will soon run the relevant agency, then write rules that lift their margins.
The Tropos story is his receipt. Twenty years ago he backed a mesh-radio startup that built free downtown Wi-Fi. Mayors loved it. Comcast and AT&T’s lobbyists got laws passed in state after state forbidding cities from competing with telecoms — sold to free-enterprise Republicans as “keeping government out of industry.” Philadelphia’s mayor went big publicly; the ex-Comcast lobbying head (who had worked for that mayor before the mayor became governor before becoming Biden’s ambassador to Canada) crushed them. “We were a sixth-grader playing Michael Jordan.”
The COVID saliva test is the cleanest case. Germany approved 85 vendors at ~$1 a test. The US FDA approved two — both run by people the relevant FDA official had worked at — at $12 a test, and then bought from them with taxpayer money. Forty-year-old commodity technology, gated to two suppliers.

Regulation is the friend of the incumbent.

He thinks the same play is now running on AI.

What he wants to fund instead: state-vs-state competition

His policy bet is on federalism as a discovery process. James Madison, he says, more or less argued this in the Federalist Papers — let states run experiments, let people migrate, follow where they migrate to. His examples are concrete and worth borrowing for any policy conversation:

Austin, housing. Rents have fallen four years running because the city actually changed zoning and building policy. “Half the country is up in arms about housing, just come look at what’s happening here.”
Houston, homelessness. Cut the homeless population by two-thirds via specific policy, well-documented, ignored elsewhere.
Pennsylvania, infrastructure. Governor Shapiro got a downed highway rebuilt in twelve days by getting bureaucracy out of the way. “Why get them out of the way just for an emergency? Get them out of the way for everything.”
Texas, school vouchers. Just passed; he’s watching whether voucher competition does to public schools what KIPP charters have already shown is possible inside the same buildings.

His Milton Friedman quote is the spine of this whole project: “One of the great mistakes is to judge policies and programs by their intentions rather than their results.”

China — what Gurley actually noticed

Gurley spends a lot of time in China and pushes back on the lazy reads. Two observations he repeats:

Of all the places I’ve visited around the world, the one place where I find the most cultural affinity between American entrepreneurs and a foreign entrepreneur is China.

He grants this is from going out to dinner and laughing rather than from a study, but he’s emphatic about it.

The second: marginal effort still buys a meaningful upgrade in living standards in China in a way it doesn’t in the US. Pulling 4-500 million people out of poverty in one generation gave a whole society a working belief that work translates to outcomes. In the US, marginal effort gets you Showtime on top of Netflix.

He cites Mike Moritz’s old FT piece — “they just work harder than we do” — and points out the cultural double standard: Americans accept a Kobe Bryant or a figure skater grinding 100 hours a week, but find the same ethic in white-collar work somehow distasteful. He doesn’t have a clean theory for why.

He also flags that one Chinese VC told him every Chinese entrepreneur reads everything coming out of American tech, and zero American entrepreneurs read anything coming out of China. “It’s not a good fact.”

Why China going open-source might be strategic, not generous

Gurley’s answer here is the most contrarian piece of the conversation, and it’s worth pulling out:

For 20 years they were accused by the West of being IP thieves. If you’re in that place where you’re accused of that, and you see Linux and you see MySQL, this looks like a pretty good solution to your problem.

The 14th Five-Year Plan reportedly singled out open source. Free their code globally and the IP-thief narrative collapses; meanwhile they get the distribution advantages open source confers (it sells itself; you sell support and add-ons on top). Gurley’s firm Benchmark has had liquidity events on more than ten open-source companies — MongoDB being the obvious public one — so he’s not theorising. He thinks open source is the closest real thing the world has to “pure competition” in the economics-textbook sense (marginal revenue equals marginal cost), which is the regime that maximises consumer surplus and which incumbents spend their lobbying budgets to prevent.

The AI bubble question

His framing is borrowed from Carlota Perez: only real waves cause bubbles. Asking “is AI a bubble?” smuggles in the assumption that bubble means fake. Real waves make people rich quickly; quick money attracts charlatans; charlatans are the bubble.

There’s real customers buying real products. There’s people taking massive bets, the biggest bets ever taken in the venture capital world. Some form of correction may not happen until five years from now, but there will be one.

For retail investors he’s almost cheerfully dismissive: own the index, you already own Nvidia, Microsoft, Google in size. Don’t try to get clever at the highest prices these companies have ever printed. Read Burton Malkiel.

The structural argument: ZIRP made everything weird

Tom asks if the K-shaped economy and risk-seeking VC behaviour both trace back to a central bank that allowed deficit spending and money printing for too long. Gurley basically agrees:

The long period people now call ZIRP, near-zero rates which was unprecedented… had that not happened, you would have had a correction in venture capital, maybe in private equity. I would have called it a healthy correction. But because of it, and then because AI came on the heels of it, that correction never happened.

He tells a Buffett story: he asked Buffett at a group function how DCFs even work at zero rates — they fall apart at negative — and whether ZIRP creates risk-seeking behaviour. Buffett’s answer: “You betcha.” Gurley says he’s “not as big of a Dalio convinced it’s all coming to an end” but also doesn’t have a confident counter-argument; he just refuses on temperament to “roll around in it if it’s inevitable.”

The pivot: why the model providers are racing to own the customer

Underneath the policy talk, Gurley drops one clean piece of competitive analysis. OpenAI and Anthropic both started by selling tokens to third parties (Anthropic powering Cursor for code). Now Anthropic has launched its own competitor to Cursor, and OpenAI is building consumer apps. Why?

A move to say I want to be directly connected to the customer is a sign — I think a pretty strong sign — that there was fear that being a model provider solely was [not] as protectable as they wanted it to be. People were switching out the model, sometimes with open source from China.

If you only sell the engine, the buyer can swap engines on you. If you own the customer relationship and the contract, you don’t have to. This is the AI equivalent of why Intel desperately wanted to be inside the laptop and why every SaaS company eventually tries to become a platform.

The book Gurley actually wanted to talk about

Running Down a Dream. He spent years reading biographies of people who started at zero and made it to the top of unconventional fields. The pattern he claims to find: continuous learning was always present, and it was always sustained by fascination, not grit. He cites Angela Duckworth’s reported revision — that if she could rewrite Grit she’d put passion above perseverance, because perseverance without passion is just suffering.

His clearest career-AI overlap: the 2023 Gallup engagement poll he keeps quoting — 29% of workers call themselves engaged, 59% are “quietly quitting.” He thinks AI is coming for the ambivalent first.

The best practices you’d learn in a physics class or an engineering class — those are in the model. The rope algorithm you had to memorise — that’s in the model. If you’re fascinated about your field, you live in the nuance, beyond those programmatic things. And that’s what’s not in the model.

His prescription for everyone, regardless of field: become the most AI-enabled version of yourself. Not as a slogan — as a survival strategy for the 71% who aren’t already doing it on weekends because they find it interesting.

Key Takeaways

Anthropic’s lobbying spend is, in Gurley’s reading, comparable to FTX’s — and the AI industry’s loud safety concerns may be cover for asking Washington to pull the ladder up.
Regulatory capture is Gurley’s master frame: regulation gets written by the regulated, especially in finance, healthcare, telecom; the Tropos and COVID-test cases are textbook examples.
His policy institute bets on state-vs-state federalism as a discovery process. Austin (housing), Houston (homelessness), Pennsylvania (infrastructure) all show targeted policy actually works; the problem is voters reward intentions, not results.
China embraced open source partly as a strategic answer to the IP-thief accusation — turn the criticism into a competitive moat.
ZIRP plus AI hitting on the heels of it produced the most risk-seeking VC behaviour Gurley has seen. The reset that should have happened in 2022 didn’t.
Carlota Perez framing on bubbles: real waves cause bubbles. AI is real and a bubble. Both can be true.
OpenAI and Anthropic moving down-stack toward consumer apps is a tell that being just a model provider isn’t defensible — buyers can swap engines, especially with Chinese open-source alternatives gaining ground.
Career advice — become the most AI-enabled version of yourself in your field. AI is coming for the 59% of workers who quietly quit; it has trouble with the artisans who live in the nuance.
Angela Duckworth, by Gurley’s recollection, has shifted to thinking passion matters more than perseverance — perseverance without passion is just suffering.
For retail investors: index funds already give you huge AI exposure. Don’t try to get clever at all-time highs.

Claude’s Take

Score: 7/10. The Bilyeu thumbnail is the worst version of itself — “Anthropic’s Hidden Money Network Will COLLAPSE OpenAI Competition” promises a forensic exposé. The actual content is one paragraph of educated speculation buried in a 105-minute meandering interview. Anyone arriving for the headline will leave annoyed; anyone willing to forget the title will get a useful Bill Gurley primer.

Gurley is at his strongest on the mechanics of regulatory capture — Tropos, COVID tests, Comcast in Pennsylvania — because he has scars and named players. He’s at his weakest when nudged into broad civilizational diagnosis, where he gets a bit “kids these days are too coddled, China just works harder.” The China observations are first-hand and worth taking seriously, but the cultural-affinity riff is a vibe, not an insight.

The most actionable nuggets for someone watching the AI stack:

The vertical integration tell. Anthropic launching a Cursor competitor isn’t strategy bravado, it’s defensive. Model providers know their position is weaker than the market thinks. Useful when valuing pure-play LLM businesses.
The lobbying frame on AI safety. You don’t have to buy the cynical read fully to notice that, empirically, every regulatory framework Anthropic has pushed for would raise costs for smaller competitors. Worth pricing in.
State-vs-state as a research lens. If you’re trying to predict where US policy will go, look at where capital and people are migrating between states. That’s where good policy actually exists, even if national media isn’t covering it.

What he gets demonstrably right: ZIRP-and-then-AI explanation for current VC excess. Probably the cleanest articulation of why nothing reset.

What he ducks: he won’t call regulatory capture corruption, won’t call himself a bear, won’t engage Dalio’s debt-cycle thesis head-on. He’s a “strong opinions, loosely held” person who in this conversation holds quite a lot of opinions loosely. Refreshing in a world of takes; not where you go for hard predictions.

The Bilyeu format hurts more than it helps — a lot of “preach!” and emotional escalation around what is, structurally, a wonky policy conversation. Find Gurley’s All-In talk on regulatory capture instead if you want the same content with no frosting.