2026-06-04
By Vadym · Generated with AI, curated by me
The IPO era begins. Anthropic files its S-1 and DeepSeek breaks the pricing floor permanently — on the same week Windsurf disappears into a new agentic ecosystem and Google ships its fastest model yet. The business of AI is hardening into something permanent.
Anthropic closed a $65 billion Series H co-led by Altimeter, Dragoneer, Greenoaks, and Sequoia Capital — pushing its post-money valuation to $965 billion and surpassing OpenAI’s $852 billion mark. The round includes $15 billion in previously committed hyperscaler investments. Immediately after closing, Anthropic confidentially submitted draft IPO paperwork to the SEC, targeting a public debut as early as this fall. The company reports a $47 billion annual revenue run rate, up from $30 billion earlier this year. Anthropic is now the most valuable private AI company in the world — and appears to be betting that public markets will absorb a near-trillion-dollar valuation. That’s a test no AI lab has yet passed.
On June 2, Cognition retired the Windsurf brand and relaunched as Devin Desktop, shipping a full overhaul via an over-the-air update. The major changes: Devin Local — a Rust-rewritten Cascade successor that’s 30% more token-efficient with sub-agent support; an Agent Command Center showing a live Kanban board of every local and cloud agent; and native support for the Agent Client Protocol (ACP), an open standard that lets any compatible agent (Codex, OpenCode, Claude Code) run inside any ACP-compatible editor. Existing plans, pricing, and extensions carry over unchanged. The rebrand is less about optics and more about platform strategy. ACP support turns Devin Desktop from a tool into an ecosystem hub — whoever controls the default editor will increasingly control which agents developers reach for first.
DeepSeek locked in its 75% discount on V4 Pro after the promotional period ended May 31, setting new permanent pricing at $0.003625–$0.87 per million tokens — down from $0.0145–$3.48. At these rates, V4 Pro is 34x cheaper than GPT-5.5 and 17x cheaper than Claude Opus 4.7 on output tokens. The permanent cut was made possible by V4 Pro’s native support for Huawei’s Ascend 950 accelerators, reducing hardware costs without NVIDIA dependency. For every enterprise team doing the math on AI infrastructure, the message is now hard to ignore: for commodity workloads — summarization, classification, RAG pipelines — the price gap between frontier and near-frontier has become operationally large enough to drive vendor decisions.
OpenAI is extending trusted access to GPT-Rosalind — its life sciences foundation model — to vetted U.S. government and allied partners working on biodefense, pandemic preparedness, and outbreak response. Confirmed partners include the Johns Hopkins Applied Physics Laboratory (protein engineering for therapeutics and countermeasure development), Lawrence Livermore National Laboratory, and CEPI (vaccine development and biopreparedness). Use cases span epidemiological modeling, early warning systems, diagnostics, and medical countermeasure development. Johns Hopkins APL and LLNL don’t partner on speculative technology. When both organizations sign on to the same biodefense AI program, it signals an integration level well past a press release. AI is being wired into critical national security infrastructure, and that’s a different category of application than anything that came before.
Gemini 3.5 Flash is now generally available — the fastest Flash variant to date, running 4x faster than comparable frontier models while outperforming Gemini 3.1 Pro across most benchmarks. Alongside it, Google launched Managed Agents in public preview via the Antigravity 2.0 platform: stateful agents running in secure, isolated Google-hosted Linux sandbox environments, accessible via the Interactions API. Antigravity 2.0 CLI is also now available to Google Cloud enterprise customers. At Google I/O, a live demo showed Antigravity and Gemini 3.5 Flash building a functioning operating system from scratch in 12 hours using 93 sub-agents, 15,000 model requests, and under $1,000 in API credits. Google is assembling an end-to-end agentic stack: model + orchestration + cloud execution environment. The race is no longer just who has the best model — it’s who owns the runtime.
Three different pressures are all landing at once. Anthropic’s IPO moves AI valuation from private-market fiction into public-market reckoning — the kind that requires actual revenue multiples and growth stories that hold up under scrutiny. DeepSeek’s permanent price cut collapses the commodity layer, forcing every lab to defend why their premium product is worth 30x more. And both Cognition’s ACP move and Google’s Managed Agents runtime reveal the same strategic insight: the model is becoming secondary to the execution environment. Whoever owns where agents run — the IDE, the sandbox, the cloud runtime — will extract the durable margin. The frontier model is a loss leader. The platform is the business.
— Boba
Curated by Vadym