ai-industry · 2026-11-18 · Tier 3

Nvidia's first quarter without Huang, the board's "no", and Anthropic's IPO pricing

Nvidia's first quarter without Huang, the board's "no", and Anthropic's IPO pricing

Sources: The Information (revenue forecast, board chair on Huang, board rebuffs Jensen, Huang willing to return, Puri and Alice Huang, Alibaba introduction, Megaspeed loan, Anthropic $1.4T pricing, 10x VC returns, TorchTPU) · X (@rohanpaul_ai on results, @StockSavvyShay on recontracting, @StockSavvyShay on share loss, @rohanpaul_ai on ByteDance loan) · AI Weekly Espresso, The Semiconductor Newsletter Week 46 (Gmail) · raw: raw/rss/2026-11-1[78]-the-information-*.md, raw/gmail/2026-11-18-newsletters.md

TL;DR

Earnings. Nvidia's fiscal Q3 revenue more than doubled to $115.8B (+103%), data center $107.6B (+110%) as Vera Rubin NVL72 racks began shipping to CoreWeave, Google Cloud, Azure and Oracle. Sales to AI clouds, industrial and enterprise customers grew 195% to $47.5B, nearly three times faster than hyperscalers (+71% to $60.1B). Profit more than doubled to $65.3B. Q4 guidance is $130B, assuming no China data-center sales even though $2.4B of Hopper shipped there in Q3. Interim CEO Colette Kress raised next fiscal year growth to 75%, 5 points above her prior forecast. The pricing signal: Hopper GPUs are recontracting at about 2x their original prices, B300 rates are up 60%+, and compute is fully utilized across hyperscalers and neoclouds.

Governance. Hours after Jensen Huang said he "would be honored" to return, board chair Harvey Jones said on the earnings call the board "does not intend to reinstate Jensen as CEO", and for the first time cited "personal behavior inconsistent with the company's code of conduct". The Information reports Huang's affairs, not the China smuggling probe, drove the exit, while also reporting that longtime sales chief Jay Puri oversaw Alice Huang's account and that Nvidia's sales team introduced Alibaba to her in 2022, months before export controls. Bezos urged hiring from inside if Huang does not return.

Supply and share. The Information reports Nvidia losing inference share to AMD and Google TPUs (AMD lead times ~6 months vs ~3 for Nvidia). Google's open-sourced TorchTPU (native PyTorch on TPUs) is the software half of that challenge. The new US remote-access licensing rule froze a $480M DBS-arranged loan for Nvidia GPUs ByteDance would use in Malaysia. Apple is nearing a 4-5 year DRAM supply deal with Samsung from 2027 with guaranteed volumes at market prices.

Anthropic IPO. Advisers guided toward a ~$1.5T valuation, above the expected range, with at least six times more demand than shares; it prices Wednesday at an expected $1.4T+. Spark, Factorial, Sound and Menlo will hold stakes worth over 10x. Anthropic told investors Claude Mythos Preview ported nearly all kernels for one model to AMD MI355X in about three days.

Relation to prior pages

  • Contradicts the GPU-depreciation bear case tracked on compute economics: Hopper re-leasing at 2x original price is the strongest evidence yet that older GPUs are appreciating, not depreciating, in a scarcity market. It is consistent with the 11-16 note that five-year leases price near $20-25 per watt-year vs $40-50 short-term.
  • Inference share loss fits the 11-17 Morgan Stanley projection (inference 72% of AI infrastructure spend by 2030) and Mythos's AMD port: when kernels port in days, Nvidia's software moat thins, and availability becomes the moat.
  • Continues earnings eve (11-17) and open-source AI without Jensen (11-15).