The Pentagon's Favorite AI Company Might Be Its Next Supply Chain Risk

The week's biggest story wasn't a funding round—it was a breakup. The Pentagon is threatening to drop Anthropic over the AI lab's refusal to let the military use Claude for "all lawful purposes." Meanwhile, humanoid robots and AI video generators are raising close to a billion each, and fusion energy just got its most credible backer yet. The capital is flowing. The question is who gets to control what it builds.

The Pentagon is considering cutting Anthropic loose. Claude was used during the U.S. military operation to capture Venezuelan President Nicolás Maduro in January—the first known use of a commercial AI model in a classified Pentagon mission. The fallout: the Pentagon is reviewing Anthropic's $200M contract and reportedly considering labeling the company a "supply chain risk" over its refusal to allow unrestricted military use. Claude is currently the only AI model on certain classified systems. More in our Feature below. Read More…

Apptronik raised $520M more for humanoid robots. The Series A extension brings total Series A funding to $935M at a $5.3B valuation—tripling its valuation from the initial close. Google, Mercedes-Benz, and B Capital co-led, with new backers AT&T Ventures, John Deere, and Qatar Investment Authority piling in. The Austin-based company is building Apollo humanoid robots for manufacturing and logistics, and recently partnered with Google DeepMind to integrate Gemini Robotics models. Nearly $1B raised before shipping at scale—that's the humanoid robotics bet in a single number. Read More…

Runway closed a $315M Series E at $5.3B. General Atlantic led, with Nvidia, Fidelity, Adobe Ventures, and AMD Ventures participating. The AI video generation company is pivoting its pitch from "make videos with AI" to "world models"—systems that simulate 3D environments with real-world physics. If it works, the applications extend well beyond media into gaming, robotics, and scientific simulation. At $5.3B, investors are betting Runway becomes infrastructure, not just a creative tool. Read More…

Anthropic has a Pentagon problem, and it's the kind that doesn't have a clean answer.

Anthropic signed a contract worth up to $200M with the Department of Defense last summer and became the first frontier AI company to deploy on classified military networks. The partnership ran through Palantir, whose software is deeply embedded across Pentagon operations. Claude was being used for intelligence analysis, document summarization, and satellite imagery processing across multiple agencies. By most accounts, it was working.

Then the Maduro raid happened. U.S. special operations forces captured the Venezuelan president in January, and according to the Wall Street Journal, Claude was deployed during the active operation. The exact role isn't clear, but the operation involved kinetic fire. People were killed.

According to a senior administration official, an Anthropic executive reached out to Palantir afterward to ask how Claude had been used—a move the Pentagon interpreted as the company potentially objecting to its product being deployed in combat. Anthropic flatly denies this, saying it has not discussed specific operations with the Pentagon or Palantir.

The deeper friction predates the raid. The Pentagon wants all AI labs to agree to an "all lawful purposes" standard—any use that complies with U.S. law is on the table. OpenAI, Google, and xAI have shown more flexibility. Anthropic has held firm on two red lines: no fully autonomous weapons, no mass domestic surveillance. The administration official described Anthropic as the most "ideological" of the AI labs, but conceded that replacing Claude would be difficult since no other model is cleared for classified systems.

The calculus for Anthropic is genuinely complicated. The safety-first brand carries weight with enterprise customers, regulators, and the engineers it needs to recruit. Caving risks internal revolt—sources say there's already disquiet among Anthropic engineers about military work. But losing a $200M contract and getting labeled a "supply chain risk" isn't good for business either.

The broader implication matters more than the specific dispute. We're watching the first real test of whether AI labs can maintain ethical boundaries while selling to the U.S. military. If Anthropic blinks, the safety-first positioning becomes marketing copy. If it holds firm and loses the contract, other labs will note the price tag—and probably conclude that principles are expensive.

The Pentagon just demonstrated something important: the government doesn't think AI companies get to decide how their tools are used in war.

Inertia Enterprises raised $450M in a Series A led by Bessemer Venture Partners, with GV, Threshold, and Modern Capital participating. The fusion energy startup was founded by ex-Twilio CEO Jeff Lawson and scientists from Lawrence Livermore's National Ignition Facility. The plan: build the world's most powerful laser, mass-manufacture fuel targets, and construct a gigawatt fusion plant by 2030. At $450M for a Series A, this is either the most expensive science project in VC history or the beginning of something real. Lawson's track record buys a lot of benefit of the doubt.

Oxide Computer closed a $200M Series C led by Thomas Tull's US Innovative Technology fund, with Eclipse, Riot Ventures, and Jane Street participating. The company builds on-prem cloud computing—integrated rack-scale systems with custom hardware and open-source software. This comes less than a year after a $100M Series B, suggesting either explosive demand or a capital-intensive hardware buildout. Probably both. As enterprises rethink cloud concentration risk, Oxide's pitch gets more compelling. Read More..

Goodfire raised $150M Series B at a $1.25B valuation, led by B Capital with Juniper, Menlo, Lightspeed, DFJ Growth, Salesforce Ventures, and Eric Schmidt. The company does AI interpretability research—reverse-engineering how neural networks make decisions. They recently identified novel Alzheimer's biomarkers by analyzing model behavior, which is the kind of result that makes a research lab look like a product company. At $1.25B, investors are betting interpretability becomes a requirement, not a nice-to-have.

Northwood Space raised $100M Series B co-led by Washington Harbour Partners and a16z, with Alpine Space Ventures and Founders Fund. The El Segundo company builds phased-array ground stations for satellite communications—hardware that can track multiple satellites simultaneously without physically moving. They also landed a $49.8M Space Force contract for satellite control network modernization. Founded by Bridgit Mendler, who went from Disney Channel to satellite infrastructure. The space ground segment is getting its venture moment. Read More...

Alva Energy launched with $33M in funding to make existing nuclear reactors produce more power, faster. The Cambridge-based startup is targeting reactor uprating—squeezing additional capacity from plants that are already built and licensed. With AI data centers creating unprecedented electricity demand, the pitch writes itself: more megawatts from existing infrastructure, no new permitting required.

Trener Robotics (formerly T-Robotics) raised $32M Series A for its AI robot skills platform. The San Francisco company builds software that teaches manufacturing robots new tasks through AI rather than manual programming. It's the picks-and-shovels play for the broader robotics boom—if every factory needs smarter robots, someone has to write the software that makes them smart.

Nucleus Security closed a $20M Series C led by Delta-v Capital for its unified vulnerability and exposure management platform. Not glamorous, but every CISO needs it. As attack surfaces expand with AI deployments, the companies that consolidate security visibility tend to get acquired at healthy multiples.

SpaceX is quietly lining up banks for a mid-2026 IPO that could raise $25B+ at a valuation approaching $1.5T. Reuters reports the company pulled in roughly $8B in EBITDA last year on $15-16B in revenue, with Starlink accounting for up to 80% of earnings. If it goes, it would be the largest VC-backed IPO in history by a wide margin. Read More...

Beta Technologies is seeking an $825M IPO at a $7.2B valuation. The electric aviation company builds charging infrastructure and electric vertical takeoff aircraft for cargo and logistics. It's a test case for whether the public markets are ready for pre-revenue deep tech hardware companies again.

Goldman Sachs acquired Industry Ventures, a secondary-focused VC manager with $7B in AUM, for up to $965M. The deal bolsters Goldman's alternatives business and signals that institutional players see venture secondaries as a structural growth market—not a temporary liquidity patch.

The Anthropic-Pentagon standoff is the one to track. Axios reports the Pentagon is close to formally labeling Anthropic a "supply chain risk," which would have cascading effects beyond just the $200M contract. Anthropic says it's having "productive conversations" with Defense, but the administration's patience appears thin. A resolution—or escalation—could come within days.

The IPO pipeline is building pressure. SpaceX's banker conversations are the headline, but Databricks, Stripe, and Anthropic itself are all in various stages of preparation for potential 2026 listings. Whether the window actually opens depends on broader market conditions, but the candidates are lining up.

On the regulatory front, watch the FTC's intensifying probe into Microsoft's cloud and AI bundling practices. Civil investigative demands went out to six or more rivals last week. If the probe expands, it could reshape how cloud giants package AI services—and how startups compete against them.