Juniper Vale: Hey — how's the week? You look like you have a thing.
Finn Brooks: I have a thing. I was at my brother's place last night and he works in defense contracting — not going to say which firm — and I mentioned I'd been reading about lobbying spend and he goes, 'yeah, we're actually cutting our Washington budget this quarter.' And I'm sitting there thinking: you are cutting it. While Anthropic just dropped $1.97 million in one quarter on federal lobbying. Alone. That's more than Nvidia spent.
Juniper Vale: Hold on — more than Nvidia?
Finn Brooks: Nvidia spent $1.25 million in Q2. Anthropic beat them by nearly $800,000. A six-year-old AI lab outspent the semiconductor giant that makes the chips powering every AI model on the planet. And it wasn't close.
Juniper Vale: So that's what we're into today — this Q2 2026 surge in AI lobbying spending, and the real question underneath it: does money moving this fast in Washington actually change who gets to shape the rules?
Finn Brooks: Right, and — okay, context, because the number gets even wilder when you stack it. Anthropic and OpenAI together spent $3.17 million in Q2 — that's a 23% jump from Q1, records for both companies. And Anthropic's first-half total already hit $3.5 million, which means they burned past their entire 2025 lobbying budget of $3.1 million with six months still left in the year.
Juniper Vale: And four years ago — no federal lobbyists. None. Not Anthropic, not OpenAI, not Nvidia.
Finn Brooks: Zero. Which is what makes my brother's thing so strange in retrospect — because the defense contractors cutting spend aren't just trimming; eight major ones collectively dropped more than 3% in Q2. The old Washington money is pulling back exactly as the labs accelerate. That is not a coincidence, I think.
Juniper Vale: But that 'not a coincidence' framing — I want to pump the brakes on it just slightly, because there's a proportionality thing that gets lost. Issue One looked at 11 major tech companies and trade associations, and collectively they spent $41 million lobbying in the first half of 2026. That's over $226,000 a day. OpenAI and Anthropic together are what, $3.17 million of that? That's like 7-8% of the total.
Finn Brooks: Okay, fair — and Oracle, which nobody is writing breathless headlines about, actually narrowly beat Anthropic. About $2 million in Q2 alone.
Juniper Vale: Right — so the Magnificent Seven are still very much in the room. The story isn't that AI labs took over Washington. It's more that they went from zero to serious players while everyone else flatlined. That's directional, not dominant.
Finn Brooks: Okay but — wait, that actually makes me want to understand what this money even does. Like mechanically. What is Anthropic buying?
Juniper Vale: Okay, so the plain version — lobbying is basically hiring someone to walk into a government office before the rule gets written and say 'here's our side of this.' That's it. It's access to the room while the room is still deciding. Federal lobbying disclosures are quarterly filings — required by law — so we can actually trace which agencies Anthropic's people visited. Department of Defense. Commerce. Treasury. That's the paper trail.
Finn Brooks: So you're paying for the conversation that happens before the vote.
Juniper Vale: Exactly. And that's — I mean, that's actually the part that matters when you're a company whose entire business model could be either protected or kneecapped depending on how export controls or AI safety standards get written. The Magnificent Seven have lawyers who've been doing this for decades. Anthropic is, what, sprinting to catch up while simultaneously trying to shape the rules from scratch.
Finn Brooks: And they're doing it fast enough that the velocity is the tell — like, first half of 2026, already past their whole 2025 budget. They're not building influence gradually. They're treating it like a product launch.
Juniper Vale: And treating it like a product launch — that's exactly what the Massachusetts thing breaks open, because that's where you stop being able to call it 'we believe in safety.' OpenAI backed the lighter Senate bill. Anthropic backed the stricter House proposal. Same state. Same moment. Both waving their safety manifestos.
Finn Brooks: No, wait — both of them simultaneously? In Massachusetts?
Juniper Vale: Same jurisdiction. OpenAI lobbying for the Senate version — safety frameworks, catastrophic risk assessments, lighter touch. Anthropic pushing the stricter House proposal. Think about what that looks like to, I don't know, a hospital compliance officer in Boston trying to figure out which AI tools she can actually bring into her department. She's watching two companies that both claim safety is their north star actively lobbying for different regulatory ceilings in her own state.
Finn Brooks: Okay and — I mean, that's the tell, right? That's the actual tell. Not the dollar amount. Because if both companies genuinely believed stricter rules were necessary, they'd be in the same room pushing the same bill.
Juniper Vale: Each one is calibrating to the ceiling it thinks it can clear more easily than its rival.
Finn Brooks: And you can see it in the disclosure documents, actually — like, look at what each company said it was lobbying on. Anthropic: export controls, cybersecurity, AI safety standards, DoD procurement, the Energy Department Genesis Mission, AI in health care. OpenAI's list is copyright, cloud infrastructure, cybersecurity, privacy. Those are not the same company's priorities dressed up differently. Those are two different business bets.
Juniper Vale: And Sam Altman is personally on the Hill briefing lawmakers while this is happening.
Finn Brooks: Which — okay, Rep. Don Beyer called the current executive order a 'Wild West environment' because it only asks for a voluntary 30-day pre-release review. Voluntary. So there actually is a vacuum, and the labs are filling it, and the part that comes later — why this is less about blocking bad rules and more about authoring the framework before a rival can — that's sharper.
Juniper Vale: Yeah — and 'Wild West' is exactly the condition you want if you're the fastest gun in the room.
Finn Brooks: And that's the lock-in piece, because — fastest gun wins, but only if the rules stick. Like, federal preemption. If Anthropic or OpenAI gets a federal framework written to their specs before open-source models scale or some third competitor emerges, that framework doesn't expire. It just... sits there, shaping everything downstream.
Juniper Vale: And it's structurally harder to dislodge than any product advantage. A rival can out-engineer you. They can't un-write a federal safety standard.
Finn Brooks: Right — and the Anthropic-DoD thing is the sharpest illustration of this, I think. They're lobbying DoD procurement officials to shape how the military buys AI. At the same time they have an active legal dispute with the Department of Defense.
Juniper Vale: Wait — suing them and lobbying them simultaneously?
Finn Brooks: Litigation and influence as complementary tools. Not contradictory. That's — I mean, that reframe kind of broke my brain a little, actually.
Juniper Vale: And export controls tie directly into that, because — okay, the China competition angle. If export controls get written broadly, they could constrain which labs can operate internationally. Written narrowly, they advantage whoever has the right architecture already deployed. That's existential, not regulatory inconvenience.
Finn Brooks: Here's the defensible version then — it's not that Anthropic and OpenAI have captured Washington. They haven't. Oracle spent more, the Magnificent Seven are still enormous. But they are the only players who have everything riding on the next 18 months of rulemaking. Microsoft is protecting an investment. These labs are playing for their existence.
Juniper Vale: That's — yeah. That's the actual calibrated claim. It's not dominance. It's asymmetric urgency. They need the framework locked before someone else does it for them.
Finn Brooks: And whoever gets there first doesn't just win the quarter. They write the floor everyone else has to build on.
Juniper Vale: The part I keep snagging on is — okay, Waymo is also in that Q2 cohort. Same record-setting quarter. And nobody is writing manifestos about Waymo's existential risk to civilization. But Bloomberg Government's numbers on Anthropic are the same data set. $3.5 million in half a year. That's real. It's just — I don't know how much of the civilizational framing is the actual lobbying strategy versus us projecting it back onto the spending.
Finn Brooks: Yeah, okay — 'Washington's new landlord' was probably a stretch. Waymo wants liability rules for robotaxis. That's normal. The difference is Anthropic is lobbying DoD procurement and export controls and health care simultaneously in one quarter, and also claiming it's the only company taking existential risk seriously. That combo is what makes it feel different. But you're right that the dollar amount alone doesn't prove it.
Juniper Vale: The real test isn't 2026 at all, I think. It's 2029, 2030 — when whatever framework these companies helped author either catches the next incident or it doesn't. That's when we find out if the safety language was vision or just a very expensive head start on writing the rules before anyone else could.
Finn Brooks: And then we'll know.
Juniper Vale: Good talk — genuinely glad you had that conversation with your brother.
