The DOJ subpoenaed the Fed over building renovations (yes, really), tech giants spent nearly $6B on energy infrastructure, and a16z raised $15B to deploy into all of it. The through-line? Everyone's betting on a future that requires unprecedented amounts of capital, power, and political cover, and traditional institutions can't keep up.

a16z raises $15B: the firm vs. fund distinction matters

Andreessen Horowitz announced $15B in fresh funds, their largest raise yet. The fundraise comes with an interesting framing: the difference between running a "fund" (optimize for carry with minimal headcount) versus building a "firm" (compound competitive advantage through platform, brand, and institutional knowledge).

It's a useful distinction as the industry consolidates. Funds get more fragile at scale, alpha decisions bottleneck through fewer people, and the next marginal deal consumes all the oxygen. Firms invest in moats: proprietary deal flow, operational support that founders actually value, media presence that shapes narratives.

The fundraise signals two things: LPs still have a massive appetite for tech exposure despite 2022-2024 markdowns, and the mega-fund model isn't dead, it's just concentrated in fewer hands. a16z now manages $90B+ across strategies, effectively operating as a financial institution. The question isn't whether they can deploy $15B; it's whether returns at that scale can match their earlier vintages when they were writing smaller checks into less crowded markets.

Worth watching: whether the "firm-building" thesis translates to differentiated returns, or whether it's sophisticated branding for what remains a scale-constrained business.

The Fed gets subpoenaed. Powell says it's about interest rates.

Federal Reserve Chairman Jerome Powell confirmed receiving grand jury subpoenas from the DOJ, nominally about headquarters renovation costs, and explicitly called them "a consequence of the Federal Reserve setting interest rates based on our best assessment of what will serve the public, rather than following the preferences of the president."

This is extraordinary. The Fed chair publicly accusing the administration of using criminal investigation as retaliation for monetary policy decisions. Markets reacted: S&P futures dipped, dollar sank, Treasury yields rose. The subpoenas came from DC US Attorney Jeanine Pirro, a close Trump ally and former Fox News host.

Whatever your politics, the institutional precedent matters. Central bank independence is a pillar of dollar credibility. If this escalates, expect currency markets to price in political risk that hasn't existed in modern American finance. The irony: tech is building infrastructure to escape slow-moving utilities (see below), but there's no private alternative to the Fed.

Tech builds its own power grid

Three deals this week tell the same story:

Combined with Stargate's broader $500B infrastructure push, tech is effectively building parallel energy infrastructure. The math is simple: AI inference at scale requires power that utilities can't deliver fast enough. So tech is buying the power plants.

This is the institutional inadequacy theme made physical. Utilities operate on regulatory timelines; AI companies operate on deployment timelines. The gap is measured in years and gigawatts. The regulatory question nobody's asking yet: at what point does "private data center" become "unregulated utility"?

Anthropic launches Claude Cowork—and needs the power to run it

Anthropic released Cowork, positioning it as "Claude Code for normies." The timing is notable: Claude Code has been having a breakthrough moment as the premier AI coding tool of choice among developers. Cowork extends that capability to non-technical users.

The strategic play is clear: if AI coding tools commoditize (and pricing suggests they will), the winner is whoever owns the broadest user relationship. But, there's a catch. Serving millions of non-technical users requires computing at a scale that makes the energy deals above look like table stakes. Anthropic's ability to secure power and chips may matter more than product quality in determining whether this bet pays off.

GPT-5.2 solves a 30-year-old Erdős math problem

OpenAI's latest model cracked a longstanding problem in mathematics, with Terence Tao confirming the result. This isn't benchmark gaming—it's genuine capability advancement in domains where verification is possible and meaningful.

The implication for AI investment: model capabilities are still improving in ways that matter, not just in ways that look good on leaderboards. That supports current valuations more than another MMLU score bump would—and helps explain why a16z is comfortable deploying $15B into this thesis.

Apple cedes AI to Google

Apple partnered with Google to power a revamped Siri with Gemini models. After the Apple Intelligence stumbles, this isn't just pragmatism; it's concession. The company that defined mobile computing admits it can't compete at the AI frontier.

For investors, this consolidates the AI race around fewer players. Apple was supposed to be a third pole in the AI competition; instead, it's becoming a distribution channel for Google. The implication: if Apple can't build frontier AI with $100B+ in annual R&D, the moat around those who can is deeper than it appeared.

The Fed story and the energy buildout are connected, even if it's not obvious. Both reflect a world where traditional institutions (central banks, utilities) can't move fast enough for the people with capital. Tech can build around slow utilities by buying power plants. But there's no private alternative to monetary policy. Whether institutional inadequacy is a feature or a bug depends on which side of the capital stack you're sitting on, and whether the institutions you can't route around are the ones under attack.