NVIDIA Puts $10B Into Anthropic's $26B Placement

NVIDIA leads a $26B Anthropic private placement, Cloudflare open-sources Kitesurf, and GitHub sets a December 14 cutoff for six Copilot models.

• Dosa AI Tools • 3 min read anthropic nvidia funding agents
In this brief · 8 sections

This brief covers AI news from 2026-11-15 UTC.

Chip money moves into the lab that buys it, an agent browser goes Apache 2.0, and Copilot starts retiring models.

NVIDIA leads $26B private placement into Anthropic

What happened: NVIDIA heads a seven-company group buying $26B of Anthropic Class N stock at the IPO price, with AMD, Broadcom, Google, Micron, Microsoft and TSMC taking the rest. The placement sits alongside an IPO seeking up to $64.3B at a valuation near $1.4 trillion.

Why it matters: The labs’ biggest suppliers are now shareholders, so compute deals and model roadmaps get negotiated with the same companies selling the chips.

Source: AIToolsRecap

Cloudflare open-sources Kitesurf agent browser under Apache 2.0

What happened: Cloudflare released Kitesurf, a stateless browser engine built for AI agents, under Apache 2.0, deployable on Cloudflare Workers.

Why it matters: Builders get a permissively licensed browser runtime they can run on Workers instead of standing up and patching headless Chrome for every agent.

Source: AIToolsRecap

GitHub sets December 14 cutoff for six Copilot models

What happened: GitHub said six models leave every Copilot experience on December 14, 2026, including Chat, inline edits, ask and agent modes, and code completions. Each has a named successor except Kimi K3, listed as to be announced; Claude Haiku 4.5 stays for individual subscribers on annual plans.

Why it matters: Pin your Copilot model choices now and check agent workflows against the replacements, or they break mid-December with no fallback for Kimi K3.

Source: GitHub Changelog

Antigravity v2.29 runs agents in the cloud with the laptop shut

What happened: Google’s Antigravity added Remote Agents and Voice Mode in v2.29, running agents in the cloud for Google AI Pro and Ultra subscribers.

Why it matters: Long agent runs stop depending on a local session staying open, which changes how you queue and hand off multi-step coding work.

Source: AIToolsRecap

Nonprofits sue five AI developers for a capability freeze

What happened: Four nonprofits filed a 63-page complaint in San Francisco Superior Court against OpenAI, Anthropic, Google, Meta and SpaceX. They seek an injunction barring systems more capable than the companies’ most advanced models as of November 12, 2026, citing California’s unfair competition law.

Why it matters: The requested order would cap frontier releases rather than pull existing products, so any team planning on a next-generation model should track the docket.

Source: Decrypt

Lenovo plans a 100 MW token factory in Zhangjiakou

What happened: Lenovo unveiled its Wanquan Hybrid Token Factory and signed with the Zhangjiakou municipal government for a first site: roughly 100 MW and a 15, 000P heterogeneous cluster in phase one. Lenovo expects over 2 quadrillion tokens a year and a 70% cut in cost per million tokens.

Why it matters: Cheaper Chinese token capacity at this scale puts pressure on per-token pricing for anyone serving inference-heavy workloads.

Source: Stillflood

Evolving Infra preprint cuts multi-agent serving latency 4.1x

What happened: Researchers from MIT, UC Berkeley and Georgia Tech posted a preprint on Evolving Infra, an open-source serving system that decouples KV cache management from GPU compute allocation for multi-agent workloads, reporting a 4.1x latency cut.

Why it matters: If you run several agents against one serving cluster, the phase-switch overhead it targets is likely part of your bill.

Source: CASRAI

Sources

Share

Discuss with AI

© 2026 dosa.dev