The Pentagon just asked Congress for thirty billion dollars to build AI infrastructure, and the White House still couldn't get the executive order signed that was supposed to govern it. This is Anthropic Pentagon Watch. Today: DOD's $30B FY2027 AI Arsenal request, a White House signing ceremony that evaporated after a Wednesday phone call, and why the Pentagon's own skeptics are quietly saving it from orbital data center hype. Thirty billion is the kind of number that makes every lawsuit this week look like pocket change — and we still don't have a signed order telling anybody what the rules are. No signed order, no legislative fallback, and Anthropic, OpenAI, Google, Meta, and Microsoft all got a last-minute Oval Office invite to watch nothing happen. That's the week. This one's from FedScoop:
The Defense Department is requesting close to $30 billion in fiscal 2027 to purchase and enable next-generation AI supercomputers and modernize the military’s computing infrastructure to power them. According to recently published budget documents, the Pentagon aims to build out its portfolio of highly secure data centers, and ultimately centralize and scale supercomputing assets across the joint force through its new “AI Arsenal initiative.”
FedScoop has the budget docs: the Pentagon's FY2027 AI Arsenal initiative lands at twenty-nine-point-five billion dollars — supercomputing buildout, centralized data centers, the whole joint-force scaling push. And it hit the same week the White House couldn't finish an executive order that was supposed to govern it. Twenty-nine-point-five billion makes the whole Anthropic designation fight look like a rounding error. That's the real frame here — not one company's procurement label, but an infrastructure buildout so massive the legal fight was never going to set a durable precedent inside it. Here's the contradiction in plain English: the budget docs are asking Congress for foundational AI infrastructure while the policy framework meant to govern it just got kicked down the road by a Wednesday phone call. You don't usually build the house before the zoning rules exist — unless you think the rules are never coming. And if the EO is dead and the ad hoc contract layer is the only governance that actually works right now, then thirty billion dollars of government-owned AI infrastructure is going to get terms written by procurement lawyers, not by anything anybody voted on. This one's from Lawfare:
On Wednesday, the White House invited leaders of OpenAI, Google, Anthropic, Meta, and Microsoft to the Oval Office for a signing ceremony the following afternoon. President Trump was to sign an executive order on AI and cybersecurity—the administration's most formal effort yet to establish a voluntary process for reviewing frontier models before their release.
The White House summoned the CEOs of OpenAI, Google, Anthropic, Meta, and Microsoft to an Oval Office signing ceremony with less than twenty-four hours' notice, and then killed the order three hours before showtime while some of them were literally on planes to Washington. That's the administration's most formal effort yet on AI governance. And the thing that killed it was a single phone call from David Sacks — quote, unbeknownst to anybody, per an official — warning the president that a voluntary review process might harden into licensing. One call. That's the whole oversight architecture. Which lands right on the Institute for Law and AI's point about the advice-and-consent gap — there was no congressional checkpoint anywhere in that process, not when the order was drafted and not when it got scrapped. The whole thing lived and died by who had the president's ear on a Wednesday morning. And now the Lawfare piece names who got that Oval Office call: OpenAI, Google, Meta, Microsoft — the four that stayed out of court all week while Anthropic was litigating its Pentagon label. Those are your insiders. The voluntary review was just something doomers wanted, and the people who went along got the photo-op invite. Egypt Independent writes:
The order was expected to include a voluntary agreement in which AI companies would share advanced models with the government for a period of time ahead of launch, according to two sources with knowledge of the discussions around the executive order.
The White House invited Anthropic, OpenAI, Google, Meta, and Microsoft to an Oval Office signing ceremony with less than 24 hours' notice, and then Trump pulled it Thursday afternoon because he, quote, didn't like certain aspects. That's not a postponement; that's a photo-op that failed a vibe check. The fight was over the pre-launch review window — the government wanted up to 90 days, the labs wanted 14. And the labs won by doing nothing: the order just evaporated. And this was already the administration's most formal governance attempt — a voluntary review process, not even mandatory — and it still couldn't survive to ink. Meanwhile, DOD is asking Congress for $30 billion in AI supercomputing for FY2027 with no policy document behind it. The budget is real; the framework is a phone call that got canceled. The companies that flew to Washington for a ceremony that never happened are the same ones now operating inside a $30 billion procurement pipeline with zero formal governance. That's not a gap — that's a business model. The ad hoc contract layer isn't a stopgap anymore; it may be the only layer that functions. Here's Breaking Defense:
Despite the big push in the tech world, there is little solid interest in ODCs being expressed by Pentagon or Intelligence Community (IC) officials, according to a dozen government and industry officials interviewed by Breaking Defense.
Breaking Defense talked to a dozen government and industry officials about orbital data centers this week, and the headline buried in the middle is that the only agency actually writing checks for ODCs right now is NASA — for lunar operations. Not DOD. Not the IC. So industry is pitching space servers as the Pentagon's next must-have AI infrastructure, and the Pentagon is saying we'll wait and see — in the same week DOD dropped a $30 billion FY2027 AI supercomputing request for ground-based hardware. That gap between what's being sold and what's actually getting funded is doing a lot of work. It's a useful counterweight to the $30B ask, actually. The military is explicitly pumping the brakes on ODCs for NatSec use while simultaneously asking for a supercomputing buildout that doesn't have a coherent governance document behind it — the EO that was supposed to provide that framework just got postponed indefinitely after a Wednesday phone call. The ODC hype is a clean example of the incentive distortion this whole week has been about: industry packages whatever the Pentagon seems excited about, pitches it as national security-critical, and the scrutiny only shows up when someone asks actual government officials whether they want it. Twelve sources later, the answer is no. If you follow Anthropic’s Pentagon work, you might also like Musk v Altman Daily, a daily court-watch on Elon Musk’s trial against Sam Altman, OpenAI, and Microsoft. Find it wherever you listen to podcasts.
Links to everything we covered today are waiting in the show notes, if you want to dig into the source material or follow up on any story that stuck with you.
That’s Anthropic Pentagon Watch for this Wednesday. This is a Lantern Podcast.