Oct 7, 2026

OpenAI at $1.4 trillion, still private. The company targets $30 billion at a $1.4 trillion valuation and keeps its IPO on hold, which means the sector's benchmark valuation will continue to be set in private markets, by a small circle of investors, rather than in a public book anyone can short. That matters for how you should read the revenue story. A private raise lets the company control its own disclosure cadence, so the question of whether recognized revenue, not bookings or an ARR run-rate, grows at the claimed rates stays open longer than it would if there were an S-1. What would strengthen the case for the valuation: audited financials, or an IPO that subjects the numbers to public scrutiny. What would weaken it: any evidence that the growth being marketed is pipeline dressed up as revenue. The related worry is circularity. If the raise is partly underwritten by compute commitments from companies whose own AI economics are thin, then one vendor's revenue depends heavily on a few customers who are themselves unprofitable on AI, and a $1.4 trillion mark is a claim about the whole chain, not one firm.

Yields are the quiet counterweight. Meanwhile Wall Street's AI-fueled boom ran into a 5.35% 10-year yield, the highest since 2002, with the comptroller warning about borrowing costs, dealmaking, and even regional tax revenues. Much of the sector's financing story assumes equity and operating cash flow carry the buildout, that hyperscaler AI capex is funded from operations rather than debt. A 5.35% risk-free rate tests that. If hyperscalers keep funding capex internally, high yields mostly hurt the marginal borrower: the neoclouds, the data center developers, the private credit stack. If they start borrowing, the math on five-year depreciation schedules against trillion-dollar revenue projections gets unforgiving fast. Evidence that would help the bull case: hyperscalers reporting positive free cash flow through this capex cycle. Evidence against: rising leverage in the data center financing layer, which today's yield environment makes more likely, and announced aggregate capex that materially exceeds what will actually be deployed, because canceled projects are the classic symptom of financing costs biting.

Open weights show up at the bank. Two quieter stories point the same direction. A Nvidia-backed startup called Reflection is readying an open-weight model for enterprise AI factories, and nearly all 50 large global banks now have responsible AI principles, with JPMorgan and Capital One setting the maturity bar. These connect. Banks are exactly the buyers who want models they can run locally, audit, and govern, which is the demand that makes open-weight models reach good enough parity for most economic use cases commercially meaningful. If Reflection ships something credible, pressure on closed API pricing follows, because API switching costs are low and customers multi-home. What would strengthen the open-weight case: independent benchmarks showing the model within striking distance of frontier leaders on enterprise tasks. What would weaken it: the usual open-weight gap on reasoning-heavy work, or banks finding integration and liability costs swamp the sovereignty savings. Either way, the governance buildout says banks plan to buy models like infrastructure, not subscriptions.

Meanwhile Modern Treasury and Rain want OCC national trust charters, and Amex is wiring card data into Perplexity, so the banks get AI and the AI companies get banks.