Prediction Markets at $20B: The House Is Betting on Itself

This week, the venture world asked three hard questions about AI: who controls it, who funds it, and who gets to bet on it. Anthropic sued the federal government to keep Claude out of autonomous weapons programs. SoftBank borrowed $40 billion to stay in the OpenAI race. And Kalshi and Polymarket each asked investors to value their platforms at $20 billion — betting that markets for everything are the next big asset class. Power, capital, and prediction. Not a bad week.

Anthropic sues the Pentagon — and the U.S. government's AI access hangs in the balance.

The DoD designated Anthropic a "supply chain risk" — a label historically reserved for foreign adversaries like Huawei — after Anthropic refused to allow Claude to be used for autonomous weapons systems and mass surveillance under its existing $200M defense contract. Trump directed all federal agencies to cease Anthropic use. The GSA terminated the "OneGov" contract the same day. Anthropic filed two federal lawsuits: one in the Northern District of California, one in the D.C. Circuit Court of Appeals, alleging First Amendment retaliation and arguing the Pentagon exceeded its statutory authority under federal procurement law.

Financial exposure is significant: hundreds of millions near-term, multiple billions if the 2026 federal pipeline evaporates. OpenAI signed its own Pentagon deal within hours of the Anthropic blacklisting, then publicly acknowledged the timing looked "sloppy and opportunistic" and is now renegotiating terms. Meanwhile, Claude surpassed ChatGPT in the iPhone App Store ranking the day after the termination — the PR battle broke differently than the contract battle. Dozens of AI researchers from OpenAI and DeepMind filed amicus briefs in support of Anthropic. Palantir, which counts Anthropic as a partner and derives 60% of revenue from government clients, moved lower on the news. This one has months to run.

Founders Fund is closing in on $6B — oversubscribed, and turning away LPs.

Peter Thiel's firm is within weeks of closing Founders Fund Growth IV at nearly $6 billion, with approximately $1.5 billion coming from the firm's own partners. The raise comes less than a year after closing the prior growth fund at $4.6 billion. In a market where LP commitment pace is slowing and mega funds are pulling back deployment, Founders Fund is doing the opposite — raising faster and larger while keeping the circle tight. The LP base is betting that Thiel's contrarian thesis — Palantir, SpaceX, Anduril, Stripe — translates to whatever the next decade of defense tech and hard infrastructure looks like. Whether that's investment conviction or brand momentum at this point is an open question. Both, probably.

SoftBank seeks a $40B bridge loan — to sit on its OpenAI position.

SoftBank is seeking the largest dollar-denominated loan in its history, underwritten by JPMorgan Chase and three other lenders at a 12-month tenor, to finance its OpenAI stake. The context: SoftBank put $30 billion into OpenAI's $110 billion round (joined by Amazon's $50B and Nvidia's $30B) at an $840 billion post-money valuation, bringing total OpenAI exposure to approximately $64.6 billion — roughly 13% ownership. S&P moved SoftBank's credit outlook to negative immediately after. The math only works if OpenAI's valuation holds or climbs. If it doesn't, Masayoshi Son has done this before — Vision Fund was underwater by $17 billion in 2019 before partially recovering. The leverage on the leverage is extraordinary.

Two years ago, prediction markets were a regulatory gray area frequented by political obsessives and degens with offshore wallets. This week, Kalshi and Polymarket are each reported to be seeking $20 billion valuations in early investor talks — nearly double what either was worth just three months ago. That jump has nothing to do with a lucky call.

The numbers are real. Kalshi crossed $1 billion to $1.5 billion in annualized revenue run rate. Polymarket processed $7.94 billion in trading volume in February; Kalshi did $9.93 billion. Combined, they moved nearly $18 billion in a single month — the GMV of a mid-sized e-commerce platform, in a market that barely existed in 2022.

The institutional money has already arrived: Kalshi's December 2025 Series E was led by Sequoia with CapitalG, a16z, and Paradigm. Polymarket's last round had ICE — the exchange giant that runs the New York Stock Exchange's parent company — investing up to $2 billion. When ICE is backing your prediction market, you are no longer a niche product.

The bull case is intellectually clean. Prediction markets are, in theory, the most efficient information aggregation mechanism available — better than expert panels, better than media consensus, better than polls. When millions of participants put real money behind outcome probabilities, the resulting prices have beaten conventional forecasts repeatedly on elections, economic data, and geopolitical events. The Iran escalation playing out this week — Brent crude above $100, WTI spiking to $118 overnight — was priced into Polymarket event contracts before most financial outlets connected the dots. That's not nothing. That's a product.

But the $20 billion valuation requires several things to go right simultaneously. First: U.S. regulatory resolution. Polymarket currently blocks American users. Its planned domestic relaunch is the core growth thesis — without it, the addressable market is capped. Legislation has been introduced to ban contracts on wars and sports events, which happen to be two of Polymarket's highest-volume categories. Insider trading allegations have surfaced. State gaming regulators are circling. The regulatory path is open but narrow.

Second: volume durability. Prediction markets spike around elections and crises. February's $18 billion combined volume is impressive — but how much of it is driven by the Iran oil shock, U.S. policy volatility, and elevated macro uncertainty? Sticky durable ARR is a different question than peak-event GMV. Kalshi's revenue figures suggest genuine monetization. Polymarket runs on crypto and doesn't publish equivalent data.

If these deals close at $20 billion each, Kalshi and Polymarket will collectively be worth more than Robinhood's current market cap, not far below Coinbase. They would be the defining data points for the sector for the next decade. The real bet isn't on the companies. It's on whether prediction markets become a mainstream financial product — the Bloomberg Terminal for event-driven risk — or stay a sophisticated niche that spikes during crises and compresses in calmer periods. At $20 billion per platform, there is no middle answer.

Ayar Labs — $500M Series E

Led by NVIDIA and AMD — a rare co-investment from direct chip competitors. The MIT spinout makes silicon-photonic interconnects that replace copper links inside AI data centers, promising up to 20x higher throughput per watt. Valuation: approximately $3.8 billion. Strategic backers including Neuberger Berman, MediaTek, and the Qatar Investment Authority also participated. When the chip giants funding your bandwidth solution are the same ones bottlenecked by it, the customer validation writes itself. March 3.

Ualá — $195M

Led by Allianz X, the Argentine neobank extended its LatAm runway with a fresh $195 million equity round. Ualá has scaled across Argentina, Mexico, and Colombia — tens of millions of accounts — making it among the few LatAm fintechs with genuine cross-border traction. Dollar-denominated investment in the region has been uneven; a European institutional anchor at this size signals continued confidence in Ualá's path to regional dominance. March 4.

Neura Robotics — ~€1B (~$1.2B)

The German humanoid robotics company is in the process of closing approximately €1 billion in new funding, anchored by Tether Holdings — yes, the stablecoin issuer. Neura makes the "4NE-1" general-purpose humanoid robot and was last valued at approximately €4 billion. Tether anchoring a German robotics round is an unusual pairing, but crypto treasury managers have been hunting for hard asset exposure and robotics fits the thesis. At €4 billion, Neura is competing globally with Figure and 1X. The field is getting crowded. March 4.

A quiet week for exits. The most notable transaction in the window was Anthropic's acquisition of Vercept, a Seattle-based desktop AI agent startup founded by Allen Institute for AI alumni (announced February 25, just outside this edition's window). Vercept's Vy desktop agent — which outperformed OpenAI and Google models on UI grounding benchmarks — will be folded into Claude's computer-use capabilities. Terms were not disclosed. The broader exit environment remains constrained: the IPO window is open but selective, and M&A is running at steady but unspectacular volume. No blockbuster exits this week.

The Anthropic-Pentagon legal proceedings are just getting started. Federal courts move faster than most assume when the plaintiff has the resources to push — watch for preliminary hearing dates and any executive branch response within the next two weeks. Congress is another variable: multiple lawmakers have already signaled interest in AI-in-weapons policy, and the Anthropic suit gives them a vehicle.

On valuations: if either Kalshi or Polymarket closes a round at or near $20 billion, expect a sector-wide reassessment of what prediction markets are worth and a rush of copycat pitches. Neither deal is confirmed; both remain in early-stage discussions. The market-moving event is the announcement, not the rumor.

The oil macro matters more than usual this week. With Brent above $100 and WTI having touched $118 overnight on Iran escalation news, any further Middle East developments would pressure late-stage valuations directly — particularly rounds currently in process that price off public comps. Nikkei down 5.2%, S&P futures off 1.1%, and Gulf supply uncertainty don't make for an easy close environment.

Founders Fund's formal close announcement is expected imminently. When it lands, watch how LPs publicly describe their thesis — whether it's a pure Thiel bet or a genuine defense/hard tech allocation will tell you something about how institutional money is thinking about the next decade.

Finally: Claude Marketplace launched this week with Snowflake, GitLab, Harvey, Replit, and Lovable as launch partners. Early traction signals — whether enterprise spend is actually routing through the marketplace or whether this is a distribution announcement without distribution — should surface in the next few weeks. Worth watching.