EAIDaily - May 22, 2026

AI English Daily — Focus: AI Coding & Embodied Intelligence Curated selection of the day’s most significant AI developments.

1. Anthropic Hits First-Ever Quarterly Profit — $10.9B Revenue, $559M Operating Profit

What happened: Anthropic projected its Q2 2026 revenue at $10.9 billion — more than double Q1’s $4.8B — and announced its first-ever quarterly operating profit of $559 million. The milestone was driven primarily by explosive growth in Claude Code enterprise adoption, with annualized revenue now exceeding $2.5B and enterprise customers spending >$1M/year doubling from 500 to 1,000+ in two months.

Why it matters: This is the first time a frontier AI lab has reached operating profitability — a watershed moment for the industry. It proves that AI coding tools are a viable standalone revenue engine at massive scale, not just a subsidized research output. For the AI coding ecosystem, it signals that Claude Code has won the enterprise developer market outright. The profitability also strengthens Anthropic’s hand in its 2026 IPO race against OpenAI and SpaceX, potentially reshaping the public-market AI landscape later this year.

Source: BuildFastWithAI · OpenTools AI

2. Boston Dynamics Unveils Production Atlas Robot — Google DeepMind Partnership Confirmed

What happened: Boston Dynamics officially unveiled the product-version Atlas humanoid robot, confirming immediate entry into manufacturing. Key specs: 56 degrees of freedom, 50kg payload, autonomous battery swapping, and operable across -20°C to 40°C. Most significantly, the company confirmed a partnership with Google DeepMind to integrate advanced AI foundation models into Atlas, giving the robot enhanced cognitive and learning capabilities. Deployments are scheduled at Hyundai Motor Group and Google in 2026, with broader customer availability in early 2027.

Why it matters: This marks the transition of humanoid robots from research prototypes to commercial-scale industrial deployment. The Google DeepMind partnership is the most concrete “embodied AI + frontier LLM” integration announced to date — combining Boston Dynamics’ world-leading locomotion with DeepMind’s general-purpose reasoning. For embodied intelligence, this is the clearest signal yet that 2026 is the year robots leave the lab and enter the factory floor. It also validates the “AI foundation model for robotics” strategy that many labs (OpenAI, Anthropic, Alibaba) are now pursuing.

Source: Boston Dynamics Blog

3. China Embodied AI Funding Hits Record $5.6B — 176 Deals by Mid-May 2026

What happened: Crunchbase data shows Chinese robotics companies raised $5.6 billion across 176 deals by mid-May 2026 — already matching the full-year 2021 peak. Key rounds: TARS Robotics ($513M seed, $1.9B valuation), Spirit AI ($435M total, “general brain” for robots), Galaxea AI ($435M total), and EngineAI ($200M Series B). The funding wave is fueling IPO momentum: Unitree Robotics has filed for a Shanghai IPO (targeting $3–7B valuation), and Robotphoenix listed in Hong Kong on May 18, closing its first day up ~80%.

Why it matters: China now accounts for >43% of global robotics VC, and embodied AI is the single largest driver. The scale and speed of funding signal that embodied AI is no longer a research curiosity — it is a fully-fledged commercial race. For the global AI community, this means the next major breakthroughs in robot learning, simulation, and deployment may come from Chinese labs and startups, not just US/European ones. The IPO momentum also creates a public-market feedback loop that will accelerate commercialization.

Source: Crunchbase News

4. Shanghai “Ge Wu” Embodied AI Platform Launches + ISO Standards Push

What happened: The National and Local Co-Built Humanoid Robotics Innovation Center in Shanghai launched “Ge Wu” (格物), an open embodied AI simulation platform. The platform’s headline claim: one codebase supports 100+ robot types without additional programming, using a general reinforcement learning framework and automated model adaptation. Separately, Shanghai announced it is pushing for a new ISO/TC299 subcommittee on humanoid robot standards, in cooperation with domestic robot companies and research institutes.

Why it matters: Simulation is the single biggest bottleneck in embodied AI — real-world robot data is scarce and dangerous to collect. A platform that can simulate 100+ robot types from one codebase is a step toward an “Isaac Sim for everyone.” The ISO standards push is equally strategic: if China defines the international safety and interoperability standards for humanoid robots, it controls the global embodied AI supply chain. This is industrial policy and AI research colliding in real time.

Source: Beijing Post · [China Daily / Xinhua coverage]

5. SpaceX, OpenAI, Anthropic: The 2026 AI IPO Race Intensifies

What happened: SpaceX formally filed its IPO prospectus on May 20 (US Eastern Time), triggering a cascade of IPO-related moves across the AI sector. OpenAI is now widely reported to be preparing a September 2026 IPO, and Anthropic — buoyed by its surprise Q2 profitability — is accelerating plans for an IPO as early as October 2026. Combined, the three companies represent a projected public-market AI valuation in excess of $2.5 trillion.

Why it matters: The simultaneous IPO of the three most important AI/space/infrastructure companies in a generation will reshape the capital structure of the entire AI industry. For AI coding specifically, public-market scrutiny will force transparency on API pricing, margin structures, and rate-limit policies — things that are currently opaque. It also means the “AI coding tools” market will soon have publicly traded pure-plays (Anthropic) alongside infrastructure plays (SpaceX/OpenAI). Developers and enterprises building on these platforms will have unprecedented visibility into the financial health of their tool providers.

Source: AI Product Hub · 36Kr

6. Gemini 3.5 Flash Outperforms Claude Opus 4.7 and GPT-5.5 on Coding Benchmarks

What happened: Independent benchmark results released post-Google I/O 2026 confirm that Gemini 3.5 Flash outperforms all competing models on 5 coding and agentic task benchmarks, including SWE-Bench variants. It achieves this at $1.50/$9 per million tokens — roughly half the price of Claude Sonnet 4.6 and a fraction of GPT-5.5’s cost. Google also confirmed that Antigravity 2.0 (its AI coding desktop environment) runs Gemini 3.5 Flash at 12× the speed of the public API.

Why it matters: For two years, Anthropic’s Claude has been the default choice for AI coding. Gemini 3.5 Flash breaking that streak — at half the price — is a genuine competitive inflection point. If Google can maintain the quality/speed/cost combination, it will force a repricing of the entire AI coding API market. For developers, it means the “best” coding model may no longer be the most expensive one. The 12× speedup in Antigravity 2.0 also hints at what is possible when the model and the IDE are co-designed — a lesson every AI lab is now rushing to learn.

Source: BuildFastWithAI Google I/O Recap · MarkTechPost

7. China AI Models Surpass 60% OpenRouter Market Share — Cost Advantage Reshapes Developer Adoption

What happened: Data from OpenRouter shows that Chinese AI models (DeepSeek V3.2, Kimi K2.6, Zhipu GLM-5.1) now account for >60% of OpenRouter API traffic, up from ~1% in 2024. The driver is overwhelmingly economic: a 10-task benchmark suite costs $4,811 on Claude, $3,357 on GPT-5.5, but $1,071 on DeepSeek V3.2 and $948 on Kimi K2.6. Anthropic’s own policy documents acknowledge that US models lead Chinese counterparts by only “a few months” and that China is winning on global adoption via cost advantage.

Why it matters: This is the first time low-cost models have overtaken frontier models on developer preference, not just price. If today’s junior developers are building on DeepSeek and Kimi, they will be the AI architects of 2027–2028 — creating a generational lock-in effect for Chinese model ecosystems. For AI coding, it means the “default” stack for the next wave of AI-native software may be built on open-weight or low-cost models, not the premium APIs from US labs. This has profound implications for pricing power, safety standards, and geopolitical AI leadership.

Source: BuildFastWithAI May 22 News

8. OpenAI Codex May 2026 Updates: Chrome Extension, CLI v0.131.0, Mobile App Expansion

What happened: OpenAI shipped a substantial Codex update in May 2026, including: (1) a Chrome extension that runs browser tasks in background tabs; (2) Codex CLI v0.131.0 with enriched TUI session controls, Vim mode, plugin marketplace support, and codex doctor diagnostics; (3) Codex mobile app support rolling out to all ChatGPT users (including free tier), with weekly active users now exceeding 4 million (8× growth since January 2026).

Why it matters: OpenAI is aggressively closing the product gap with Anthropic’s Claude Code. The Chrome extension and mobile app push Codex from a “developer tool” to a cross-platform AI coding substrate — available wherever the user is. The 8× WAUs growth confirms that AI coding is expanding beyond professional developers to students, analysts, and power users. The CLI v0.131.0 improvements (Vim mode, plugin marketplace, diagnostics) also signal that OpenAI is building for serious long-term users, not just casual trial. The AI coding tools market is now a three-way race: Claude Code (enterprise), Codex (breadth), and Gemini/Antigravity (speed/cost).

Source: ReleaseBot Codex Changelog · Authority AI Tools · 36Kr

Summary: The Big Picture

Report compiled: May 22, 2026 07:36 GMT+8 Next scheduled run: May 23, 2026 07:36 GMT+8