Capital Week in AI: Big Rounds, Bigger Capex, and a Probabilistic Computer

The roundup: rounds, capex, and policy

Mayfield's discovery roundup collects a week of AI capital and policy items that are large by any measure: OpenAI is reportedly seeking up to 30 billion dollars in new funding, Anthropic and Nscale are reportedly negotiating a 45 billion dollar AI computing agreement, and Meta expects 2026 capital expenditure of between 135 billion and 145 billion dollars. m:…4783

The roundup also records OpenAI acquiring Orchestra Bio to expand its drug-discovery capabilities, the U.S. ordering federal civilian agencies to remove DeepSeek AI products from government systems, and Anthropic proposing federal AI safety legislation that could restrict models posing imminent catastrophic risks, which puts corporate strategy and government action in the same weekly frame. m:…4783

Taken together, the items describe a market where the largest checks are going to compute and to capability, while governments are simultaneously setting removal orders and debating safety legislation, so capital allocation and policy are moving on parallel tracks rather than in sequence. m:…4783

Horowitz: what changes for founders and VCs

Ben Horowitz, speaking on the a16z podcast, framed the technology shift directly: AI is a probabilistic computer, whereas previous computing eras were based on deterministic computers, and AI applications can reach massive user numbers instantly because they do not rely on network-effect adoption curves. m:…2210

His advice on advantage follows from that framing: durable advantage in AI applications comes from proprietary data and customer possession rather than just model architecture, successful market entry requires a product 10 times better than the incumbent rather than marginally better, and entrepreneurship requires a mindset where the founder feels compelled to build the company. m:…2210

Horowitz also described how the venture industry itself has changed, evolving toward hiring former operators to provide more relevant advice to founders, with the primary role of a venture capitalist being to provide momentum in hiring, fundraising, and customer acquisition, and he predicted AI video and entertainment will emerge as a new storytelling medium distinct from film and theater. m:…2210

The inflation warning

Bloomberg's market podcast adds the macroeconomic counterweight: structural spending drivers, including AI, create an inflationary impulse that keeps interest rates high for a long time, which means the same buildout driving the funding rounds may also keep financing costs elevated. m:…4735

The podcast's other items show AI reaching retail finance directly: Robinhood has launched agentic trading accounts allowing retail customers to direct AI to manage portfolios, a step that moves AI from a market theme into a product feature inside brokerage accounts. m:…4735

For investors, the three sources together suggest a discipline: fund the data and customer advantages Horowitz describes, watch the capex and compute agreements Mayfield tracks, and price in the possibility Bloomberg raises that AI-driven spending keeps rates higher for longer. m:…4783 m:…2210 m:…4735

A discipline for the capital week

The capital week's lesson is selectivity: Horowitz's test of proprietary data and customer possession, plus a product ten times better than the incumbent, gives investors a filter for the large rounds and compute agreements Mayfield tracks, while Bloomberg's inflation warning adds a cost-of-capital check on the same deals. m:…2210 m:…4783 m:…4735

Policy belongs in the same diligence: removal orders for DeepSeek products in federal systems and proposed safety legislation that could restrict high-risk models show that the regulatory environment can change the market for a model as surely as a funding round can. m:…4783