Ten Days After the Slowdown Call, the Model Layer Starts a Price War; Alibaba Tables a Full-Stack Compute Timeline 48 Hours Before the Summit; Muse Writes Its First Victim List

1. The slowdown call survived ten days: OpenAI and Anthropic open a token price war 90 minutes apart — the inference cost curve resets 20-50% overnight

What happened.During Tuesday's US session, Anthropic shipped Claude Opus 5.5 first: standard API at $4/$20 per million input/output tokens — 20% below Opus 5 — with "Fable 5.1-level performance at ~40% lower cost in typical workloads," and a Fast mode at 2.5x speed ($8/$40); live on AWS, Google Cloud and Azure. Ninety minutes later, OpenAI released GPT-6 Sol and Luna: Sol at $2/$10 and Luna at $0.10/$0.50 — straight 50% cuts versus the prior generation and below GPT-5.6 promotional pricing; cached reads now 90% off, with GitHub measuring a >50% drop in tokens needing fresh processing. Altman: "half the price per token, and even less per task."

Why it matters.The timing is the whole story: on 9/12 Amodei posted that "we must slow the pace at which we improve the capabilities of AI models"; on 9/22 both labs shipped bigger-and-cheaper on the same afternoon — The Register counts Anthropic's cadence accelerating from roughly quarterly in 2025 to nearly monthly in 2026. The antitrust suit's core claim (coordinated slowdown) did not survive ten days of commercial pressure. Anthropic also published a transparency report disclosing that "Claude leads 26%" of its own R&D — reading as a pre-emptive offering to regulators and the court. The price war transmits two ways: OpenAI/Anthropic unit revenue compresses (against a reported $278B cumulative burn, volume must now cover price), while agent unit economics improve directly — Muse-class "one VM per user" workloads are the biggest beneficiary of cheaper tokens.

Investment implication.Transmission: token prices -50% → per-task agent cost down → agent penetration accelerates → a steeper second curve for server CPU/DRAM/SSD and inference compute — two ends of the same chain as this week's CPU/memory rally. Watch whether MU's 9/30 call is the first to cite agent workloads as a server-memory demand driver. Kill condition: if token consumption fails to volume-up after the cuts, the price war stops being "deflation for penetration" and becomes "cutting your own ARPU."

HIGH IMPACT Confidence: HIGH (both official releases; first-party pricing) Affected: MSFT US · GOOGL US · AMZN US · META US · ORCL US · CRWV US · NVDA US