ONE THING TO KNOW
Anthropic raises $65B, with three memory suppliers are in the round.
Anthropic closed a $65 billion Series H at a $965 billion post-money valuation — the largest private funding round on record, overtaking OpenAI as the most valuable AI startup — led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with Samsung, SK Hynix, and Micron joining as strategic infrastructure partners. The same three companies supply the HBM inside the GPU clusters Anthropic is paying $1.25 billion per month to operate via SpaceX, alongside a $100 billion 10-year arrangement with AWS Trainium and a $200 billion five-year commitment to Google TPUs beginning 2027. The round closed with Anthropic’s annualized revenue run-rate crossing $47 billion this month, and is expected to be the company’s final private raise before an October 2026 IPO.
Simultaneously, Anthropic launched Claude Opus 4.8 — scoring 69.2% on SWE-Bench Pro, ahead of GPT-5.5 and Gemini 3.1 Pro — at 3× lower cost than its predecessor. The model introduces Dynamic workflows in Claude Code, enabling parallel subagents for codebase-scale migrations. Anthropic also teased Claude Mythos, described as more capable than Opus 4.8, arriving “in the coming weeks.”
PUBLIC MARKETS
Q1 FY2027 results tonight: $2.418B revenue (+28% YoY), record data center of $1.83B, guiding Q2 at $2.7B midpoint (+35% YoY — an acceleration). The number to carry: Marvell now has 18 active custom silicon projects, six for companies outside the four named hyperscalers — the first public signal that the bespoke AI accelerator market is extending to a second tier of operators. Custom silicon projected to more than double in FY2028. Stock up approximately 7% in after-hours to ~$222.
Qualcomm gave back most of yesterday’s 11.6% ByteDance-driven surge (ATH close of $248.82) as investors took profits today. The deal itself, reported by Bloomberg, is structurally important: Qualcomm will supply ASICs and chip manufacturing services to ByteDance for AI data centers, operating within US export control limits. ByteDance is Qualcomm’s first major ASIC customer for data center AI — the push from smartphone processors into custom silicon for inference is now a signed contract, not a roadmap slide.
PRIVATE COMPANIES
FuriosaAIpartnership
Third-generation AI inference chip announced yesterday in co-development with Broadcom, using Broadcom’s XDSiP chiplet packaging, 2nm fabrication, and 12 HBM4/HBM4e stacks (432 GB). Targets inference clusters scaling to thousands of nodes via Broadcom Ethernet fabric, removing the current 8-chip ceiling. Sampling H1 2028.
THE COMPUTE 100 PODCAST
THIS WEEK
Erik Bernhardsson, CEO of Modal
Fresh off Modal’s Series C, Erik joins Brian and Gaby for this week’s episode. Modal raised $355M at a $4.65B valuation — up from $1.1B at Series B — with annualized revenue at ~$300M, driven by surging demand for its sandboxes product, which lets AI agents run code in isolated environments. Founded in 2021 to reduce developer friction around deploying ML models at scale, Modal has grown to become the go-to serverless inference platform for teams building with LLMs and agents.
Listen on Spotify ↗