Anthropic nears a $30B close at $900B — briefly the most valuable private AI company on Earth
Dario Amodei's revenue run-rate hit $30B ARR in April; as the round closes this week, OpenAI's $5.7B Q1 and Google's I/O multimodal push keep every other scoreboard moving too.
The story
The number that matters most this week isn't a benchmark — it's $30 billion. Anthropic is set to close its latest funding round at that figure, at a pre-money valuation above $900 billion, according to Bloomberg's May 22 report. That would push Anthropic past OpenAI's $852 billion March valuation, making it, on paper, the most valuable private AI company in the world. The fuel for that claim is real revenue: Anthropic hit $30 billion in annualized run-rate in April 2026, up from $87 million in January 2024. Dario Amodei has said the trajectory beat internal forecasts by a factor of eight. Sequoia, Dragoneer, Altimeter, and Greenoaks are each committing roughly $2 billion, with Founders Fund and General Catalyst also in. None of this money is for marketing — Anthropic is building compute and hiring to sustain the Claude 4.x revenue cycle. The next move is OpenAI's: the company that still generates more quarterly revenue ($5.7B in Q1) now needs to answer the valuation headline without overpaying on its own next raise.
Who moved this cycle
| Player | What changed | Consequence |
|---|---|---|
| Anthropic | Bloomberg reported May 22 that a $30B+ round co-led by Sequoia, Dragoneer, Altimeter, and Greenoaks (~$2B each) is set to close as soon as this week at a >$900B pre-money valuation, eclipsing OpenAI's $852B March mark · April ARR hit $30B, up from $87M in January 2024 | If the close confirms, Anthropic briefly holds the most-valuable-private-AI-company title and gains capital to fund frontier training without the equity-for-cloud-credits structures that constrain most labs |
| At Google I/O 2026 (May 19), released Gemini 3.5 Flash for agentic/coding workloads, Gemini Omni for native video generation from any input modality, and Gemini Spark — a 24/7 personal AI agent rolling into the Gemini app | Three simultaneous product angles — developer API, consumer video, ambient agent — is a breadth play: Google bets no single rival holds all three at once, making displacement harder | |
| OpenAI | GPT-5.5 launched at $5/$30 per 1M tokens and became ChatGPT's default via GPT-5.5 Instant · Launched the OpenAI Deployment Company with $4B initial investment and Bain & Company as a founding partner | The Deployment Company adds a systems-integrator layer beneath the API — enterprise buyers get hands-on support, directly eroding Accenture and Deloitte AI practice positioning |
| xAI | Grok Build 0.1 launched May 14 as a purpose-built coding agent (256K context, SuperGrok Heavy only at $300/mo) · Third-party connectors added May 22: Vercel, Canva, Gamma, S&P Global · Apollo Global Management and Morgan Stanley began internal Grok testing | Wall Street on-boarding signals xAI is targeting OpenAI's enterprise finance vertical; a win in financial research workflows gives xAI an enterprise moat independent of consumer mindshare |
| Cohere | Released Command A+ on May 20 under Apache 2.0 — first major Command-family model commercially usable without a separate Cohere license · Signed MoU with Indra Group (Spain's defence prime) for sovereign AI covering Spain, Canada, and Europe | Apache 2.0 invites self-hosting at regulated enterprise scale; the Indra deal opens a defence revenue line that US-based labs can't easily enter under current export rules |
| Alibaba | Unveiled Qwen3.7-Max at Alibaba Cloud Summit in Hangzhou — designed for agentic tasks with up to 35 hours of autonomous operation · Integrated Qwen into the Taobao app (May 11) with access to 4 billion Taobao and Tmall products | Taobao integration gives Qwen a high-frequency consumer deployment at scale; the usage signal from billions of shopping queries will compress Qwen's next fine-tuning cycle significantly |
New models on the board
| Player | Model & what changed | Pricing |
|---|---|---|
| Anthropic | Claude Opus 4.7 Released April 16 · SWE-bench Pro jumped 10.9 points from Opus 4.6's 53.4% to 64.3% — biggest single-version gain of 2026 · improved vision resolution · new tokenizer (up to 35% more tokens per prompt vs 4.6) | $5/$25 per 1M tokens (unchanged from Opus 4.6) |
| OpenAI | GPT-5.5 / GPT-5.5 Instant New flagship with perfect AIME 2026 math score (company-announced) · GPT-5.5 Instant becomes ChatGPT's default model · GPT-5.5 Thinking is default for ChatGPT with connected financial accounts | $5/$30 per 1M tokens; cached input at $0.50/M |
| Gemini 3.5 Flash / Gemini Omni 3.5 Flash: frontier performance for agents and coding, outperforms Gemini 3.1 Pro on key benchmarks (company-announced) · Omni: native video generation blending text, audio, image, and video inputs | Pricing not yet published; expected within days of I/O | |
| xAI | Grok 4.3 Launched May 4 · 1M-token context · native video input · built-in reasoning · xAI calls it their fastest, most intelligent model to date (company-announced) | API pricing not publicly listed |
| DeepSeek | DeepSeek V4-Pro / V4-Flash Released April 23-24 · MIT license with downloadable weights · strong benchmark performance with pricing well below US lab equivalents | Significantly below US lab tiers; exact rates unconfirmed by multiple independent sources |
Market signals
| Player | Type | Value | Context |
|---|---|---|---|
| Anthropic | Funding | $30B+ round at >$900B valuation (Bloomberg, May 22 — unconfirmed, no term sheet signed) | Would surpass OpenAI's $852B March valuation; Sequoia, Dragoneer, Altimeter, Greenoaks each ~$2B; Founders Fund and General Catalyst also participating |
| Anthropic | Revenue | $30B ARR as of April 2026 (company-announced) | Up from $87M in January 2024 — 340x in 27 months; Dario Amodei said it beat internal forecasts by 8x |
| OpenAI | Revenue | $5.7B Q1 2026 quarterly revenue | Enterprise now >40% of revenue; Deployment Company targets enterprise-to-consumer parity by end of 2026 |
| OpenAI | Partnership | OpenAI Deployment Company launched May 11 with $4B initial investment, Bain & Company as founding investor-partner | 19 global partners at launch; hires hundreds of Forward Deployed Engineers to embed inside enterprise clients |
| Thinking Machines Lab | Partnership | Multi-billion dollar Google Cloud deal for GB300 compute (TechCrunch, April 22) | Mira Murati's lab secures infrastructure at scale before any public model ships — signals aggressive training roadmap in 2026 |
| Cohere | Funding | ~$20B combined valuation post Aleph Alpha merger; $600M Schwarz Group Series E anchor (company-announced) | Creates a European-rooted AI firm covering sovereign AI, defence, retail, and government verticals across the EU |
| xAI | Customer | Apollo Global Management and Morgan Stanley began internal Grok testing (May 2026) | Elon Musk joined Trump's China state visit alongside Jensen Huang — xAI positioned to capitalize on any US-China data or chip deal momentum |
What builders are saying
Andrej Karpathy (Anthropic, formerly OpenAI founding team). Joined Anthropic to work on using Claude to improve Claude itself — the most prominent cross-lab talent move of 2026, signaling that the frontier research competition has intensified beyond what any single lab can staff from within
Simon Willison (simonwillison.net, PyCon US 2026 lightning talk). Delivered a whirlwind six-month LLM recap at PyCon US 2026, noting laptop-available models now wildly outperform expectations from a year ago, and calling prompt injection the persistent under-solved security problem in agent deployments — 'a bit like the wild west of early computing'
lmcouncil.ai / Chatbot Arena (May 17, 2026 — 6.3M votes). Claude Opus 4.6 hit a coding Elo of 1561 — the first model to break 1500 in Arena history; GPT-5 leads on math (perfect AIME 2026); no single model wins all categories, making model choice a cost and integration decision rather than a raw capability one
Air Street Press (State of AI: May 2026). Named April–May 2026 the most competitive window in AI history: GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro all within benchmark noise of each other, with more frontier model launches in 22 days than in any prior month on record
Threads worth pulling
Anthropic $30B ARR → revenue covers training costs from operations → reduced dependence on AWS/Google cloud credit deals → renegotiation leverage on next infrastructure contract Labs that self-fund compute from revenue no longer need equity-for-credits structures. At $30B ARR with strong token margins, Anthropic approaches the threshold where its next GPU cluster can be owned, not rented — the same dynamic that gave Google DeepMind independence inside Alphabet.
OpenAI Deployment Company → competes directly with Accenture and Deloitte AI practices for enterprise budgets → system integrators respond by deepening model-level customization → fine-tuning demand rises across all labs When OpenAI enters the professional services layer, incumbents differentiate on customization. That work runs on API tokens — a net positive for Anthropic and Google's enterprise tiers, even as OpenAI wins the headline deals.
Trump lifts Nvidia H200 chip ban on China → Alibaba, Baidu, DeepSeek get better hardware faster → training costs fall for Chinese labs → Qwen and DeepSeek pricing benchmarks drop further → US API pricing faces renewed pressure The chip ban's removal is a geopolitical event with a model quality downstream. Better hardware in China means the next DeepSeek and Qwen iterations are cheaper and faster to train — and they already price at a fraction of US equivalents.
Cohere Command A+ Apache 2.0 released three months before EU AI Act full applicability → European enterprises self-host to reduce GPAI compliance burden → Cohere's Aleph Alpha merger positions it as the default sovereign AI vendor for EU regulated sectors The EU AI Act's obligations hit third-party API consumers harder than self-hosters on data governance requirements. Cohere releasing an Apache-licensed model ahead of the August 2 deadline is a well-timed product move for European procurement cycles.
Gemini Omni native video generation inside the Google ecosystem → YouTube Creator Studio gains AI video tools without third-party friction → Runway, Pika, and Sora face creator budget competition from a free/cheap platform-native tool Omni doesn't need to be better than Runway — just good enough and already inside the workflow for 100M+ creators. Platform-native AI compounds when distribution is free and switching cost is zero.
What we’ll be watching
- Anthropic $30B round close — Bloomberg said 'as soon as this week' on May 22; watch for official announcement from Dario Amodei, likely paired with a revenue milestone disclosure
- OpenAI formal revenue statement — the $5.7B Q1 figure needs confirmation from a second independent source; watch for a press round or SEC-equivalent disclosure tied to the Deployment Company launch
- Gemini 3.5 Flash and Omni API pricing — Google announced models at I/O but hasn't published developer token costs; pricing expected within days and will set the new agentic benchmark floor
- Nvidia China H200 sales restart — Jensen Huang's presence with Trump in China positions a formal announcement; watch for Alibaba and Tencent as first named customers, which would signal scale of demand
- EU AI Act final compliance guidance — full applicability is August 2, 2026; expect major labs to publish compliance documentation in June, with Mistral and Cohere likely first given European headquarters
Sources: bloomberg.com, techcrunch.com, techtimes.com, anthropic.com, openai.com, scmp.com, engadget.com, ai.google.dev, sacra.com, press.airstreet.com, lmcouncil.ai, simonwillison.net, cnbc.com, ciodive.com, b2bnn.com.