The Investor

CoreWeave fell 15.5% on Oct 14 after hot inflation data. In the same stretch, AirTrunk's $1.5B REIT and Firmus's $30.6B listing were pulled, The Information reports. Every stated objection was rates, leverage or a thin operating record, not weak demand. For you, that makes cost of capital the first variable in any compute deal. Fortune's Term Sheet quotes Goldman's David Solomon describing not a closed window but a 10-15% price gap that issuers won't accept.

Eight state attorneys general sued OpenAI on Wednesday under state consumer-protection law, Casey Newton reports. Meta lost $942M in New Mexico on the same legal theory. The theory sidesteps Section 230 by targeting what a company said publicly about safety. For you, consumer-chatbot liability belongs in the model as a cost line, not a tail risk. A New York-led 42-state probe is still pending, so these suits look like the first rung.

Workday agreed to buy agentic identity platform C1.ai, per TLDR IT. Separately, Anthropic struck a license-and-hire deal with Opaque Systems, The Information reports. Neither deal disclosed terms. JPMorgan is also blocking outside agents like Meta's Muse from client accounts. For you, the control layer now has buyers, but the exits so far are tuck-ins and license-and-hires. Defense-trained models still fall to prompt injection 32-58% of the time, per Techpresso. Because the model cannot reliably police itself, controls have to sit outside it, which is why systems of record and labs want to own that layer rather than rent it.

ASML says its supply plans fall short of demand for 2027 and 2028, and some customers are asking to place orders into 2029-30, per Techpresso. ASML also raised the low end of its 2026 sales guidance to €44B-€45B. Microsoft will make its Maia chips the main inference system for its own products at multi-gigawatt scale from 2027, per TLDR Hardware. For you, scarcity sits upstream in lithography and power gear. Once Maia takes over, those Microsoft workloads will no longer draw on merchant GPUs, which caps what Microsoft's product inference can add to merchant-GPU demand.

a16z's David Haber argues that long-running agents can own weeks-long customer journeys. If they do, vendors can price on outcomes, and AI's market grows from a share of labor budgets to a share of revenue. His proof point is Red Ventures, which is paid per closed insurance policy. Its EBITDA fell from about $370M in 2022 to an estimated $160M in 2025. That undercuts the example more than it supports it: being paid per outcome did not protect earnings, and the post does not explain the decline. For you, outcome-linked revenue needs its own quality discount, and the post includes no contract or valuation data.