AI Daily 2026-05-28 | Anthropic Overtakes OpenAI, SpaceX Exposes Orbital AI Empire, Huang Renxun Locks Rubin Capacity in Taiwan

Three things are happening simultaneously: Anthropic’s enterprise adoption rate surpassed OpenAI for the first time (34.4% vs 32.3%), and Claude Code’s market share reached 54%; SpaceX disclosed IPO documents, where the core narrative of its $1.75T valuation shifted toward orbital AI data centers; Jensen Huang arrived in Taiwan to lock in Vera Rubin mass production commitments with TSMC, as NVIDIA’s annual spending in Taiwan surged from $150 billion to $1500 billion. The common signal: Control of AI is shifting from the model layer to the physical layer—whoever controls the computing power infrastructure holds the ticket to the second half of the game.

📖 This Issue’s Watch List In-depth Guide Link to heading

1. 🔥 The Official Handover of Enterprise AI Power — Anthropic Fully Surpasses OpenAI Link to heading

In May 2026, two independent sets of data simultaneously confirmed a historic moment.

Ramp AI Index (based on real payment data from 50,000 enterprises): Anthropic’s enterprise adoption rate reached 34.4% (+3.8% MoM), surpassing OpenAI’s 32.3% (-2.9% MoM) for the first time. 12 months ago, Anthropic was under 8%, representing a fourfold increase; OpenAI grew by only 0.3% over the same period. Ara Kharazian, Chief Economist at Ramp, stated: “This is a stunning reversal in the competitive landscape of AI model providers.”

On the same day, Anthropic’s CFO disclosed: The annualized revenue run rate broke through $30 billion, surpassing OpenAI’s $24 billion. It took only 4 months to jump from $9 billion to $30 billion. More than 1,000 enterprises are paying $1 million+ annually, doubling since February. Dario Amodei said at the Code with Claude conference: “We originally planned for 10x growth; we actually hit 80x.”

The growth engine is crystal clear — Claude Code. Menlo Ventures data shows it holds a 54% market share in AI programming tools; a JetBrains April survey showed 46% of developers listed it as their “favorite tool” (Cursor 19%, Copilot 9%). Developers use it for an average of 20 hours per week. The growth model is bottom-up individual paid penetration — exactly like Dropbox and Slack.

Meanwhile, OpenAI announced the formation of The OpenAI Deployment Company (DeployCo) — led by TPG with 19 PE firms participating, starting with $4 billion in capital, and OpenAI itself committing $1.5 billion. Combined with independent partnerships with Accenture and Cognizant, they are taking the “capital + system integrator” route. The contrast between these two paths answers the same question: AI is no longer a game of who has the best model — it’s about who can make customers unable to live without them.

(Ramp, Axios, VentureBeat, CNBC, Menlo Ventures, The Pragmatic Engineer)

2. 🚀 SpaceX S-1 Exposed: The AI Orbital Compute Bet Behind the $1.75T Valuation Link to heading

SpaceX has released its public IPO filings. Core numbers:

• Valuation target $1.75 trillion (surpassing Meta, approaching Amazon)
• 2025 revenue $18.7 billion, operating loss $2.6 billion
• Musk holds the triple roles of CEO, CTO, and Chairman, with personal holdings of approximately $800 billion
• Anthropic pays SpaceX Colossus 1 $1.25 billion per month in compute fees, with a contract running until May 2029 (totaling over $40 billion)

What excites the market most is the shift in the IPO’s core narrative: from the Starlink communication network to orbital AI data centers. The S-1 proposes deploying 100GW of compute power to Sun-synchronous orbit annually, starting in 2028, with long-term goals including TW-level orbital compute and a lunar AI energy hub. The S-1 estimates the Total Addressable Market (TAM) for AI orbital computing at $28.5 trillion.

Ben Thompson of Stratechery pointed out: SpaceX’s core advantage isn’t the rocket, it’s the launch cost — which has dropped to less than 1/10 of traditional prices. When the cost per kilogram to orbit is this low, “building data centers outside the gravity well, using free solar energy and vacuum cooling” moves from science fiction to a capital narrative.

Two major risks: First, Musk’s triple role as CEO/CTO/Chairman creates a governance time bomb regarding the conflict of interest between xAI and Anthropic (SpaceX’s largest AI customer). Second, maintaining 100GW of orbital infrastructure with “thousands of launches” per year at Starship frequency requires unprecedented launch reliability.

(TechCrunch, Bloomberg, Reuters, Stratechery, SpaceX S-1)

3. 🇹🇼 Jensen Huang Arrives in Taiwan: Vera Rubin Mass Production Locked and “The Largest Product Launch in Taiwan’s History” Link to heading

NVIDIA CEO Jensen Huang arrived in Taipei on May 27, beginning a two-week visit ahead of COMPUTEX 2026. Core agenda:

• Locking in Vera Rubin Production Capacity at TSMC: Negotiating mass production commitments for the next-generation Rubin platform with TSMC Chairman C.C. Wei. Huang called it “probably the largest product launch in the history of Taiwan.”
• Groundbreaking for Beishike Headquarters: Groundbreaking ceremony on 5/27, attended by Taipei Mayor Chiang Wan-an.
• COMPUTEX Joint Keynote: Will take the stage with Marvell CEO Matt Murphy, with the theme “The Future of AI Depends on Connectivity” — hinting that network architecture will become the next battlefield after GPUs.
• Employee Town Hall: Huang disclosed that NVIDIA’s annual spending in Taiwan has surged from $10-15 billion 4-5 years ago to $150 billion, and personally appealed to Chiang Wan-an for increased power supply.

An undercurrent to watch: Three days before Huang’s arrival, Taiwanese prosecutors launched an investigation into three suspects suspected of using forged documents to illegally export Super Micro AI servers containing NVIDIA chips to mainland China. NVIDIA’s market share in the Chinese AI accelerator market has plummeted from 95% to near zero. Huang publicly stated in April that export controls “have largely backfired.”

(TechTimes, Focus Taiwan, NVIDIA Official Blog, Tom’s Hardware)

4. 🏢 New Milestones in Enterprise AI Deployment Link to heading

• KPMG deploys Claude to 276,000 employees — one of the largest single enterprise AI deployments to date, marking the transition of the Big Four accounting firms from the experimental stage to the regulatory infrastructure level.
• Canada rules ChatGPT violates PIPEDA privacy law — the first national-level privacy ruling by a G7 nation against a general-purpose AI chatbot, which could serve as a global template.
• Cohere merges with Aleph Alpha to create a $6.8 billion “Sovereign AI” European alternative.
• OpenAI internal model independently proves the 80-year-old unsolved Erdős conjecture — a major independent discovery by AI in the field of pure mathematics.

🌐 X Platform AI Hot News Briefs Link to heading

Topic 1: Anthropic Surpasses OpenAI Sparks Intense Debate in Developer Community Link to heading

Following the release of Ramp data, the developer community has engaged in a heated discussion over whether “Claude Code’s 54% market share is sustainable.” Supporters believe bottom-up product penetration (individual developer payment → entry into enterprise) is more stable than Top-down enterprise sales; skeptics point out that a 3x increase in API prices and Colossus’s monthly $12.5 亿 computing bill are ticking time bombs.

Topic 2: Musk’s Triple Role Sparks IPO Governance Controversy Link to heading

SpaceX S-1 confirms Musk will serve as CEO, CTO, and Chairman, with personal holdings of approximately $8000 亿. The overlap of interests between xAI and Anthropic (SpaceX’s largest AI customer at $12.5 亿 monthly) has been flagged by multiple analysts as “the largest unpriced risk of the IPO.” Stratechery commented: “The biggest problem with the largest IPO in history isn’t the business model—it’s the governance.”

Topic 3: G7’s First AI Privacy Ruling—Canada vs. ChatGPT Link to heading

The Office of the Privacy Commissioner of Canada ruled that ChatGPT violated PIPEDA, with issues focusing on the processing of personal information in training data and the right to erasure for users. Regulators in Europe and Japan have indicated they will closely monitor the details of the ruling. This could be the most far-reaching regulatory action against the AI industry since the EU AI Act.

Topic 4: Erdős Conjecture Solved by AI—A Watershed Moment for Scientific Reasoning Link to heading

An internal OpenAI model has independently proven a geometric conjecture proposed by Paul Erdős in 1946. This marks the first time AI has made an independent discovery on a pure mathematics problem with a difficulty level above top-tier human mathematics competitions. The arXiv preprint has already sparked intense discussion in the mathematics community—some scholars call it “a qualitative leap in AI’s scientific reasoning capabilities,” while cautious researchers point out that the readability of the proof remains questionable.

Topic 5: KPMG Fully Deploys Claude for 27.6 万 Employees Link to heading

This is not only an enterprise customer milestone for Anthropic. As the Big Four accounting firms begin to have all staff use AI tools, AI-assisted work has shifted from “efficiency improvement” to a “compliance requirement”—not using AI will no longer be an efficiency disadvantage, but an audit risk.

📊 Key Data Link to heading

🧠 Deep Observation Link to heading

May 28 is no ordinary day. Three unrelated events occurred on the same day, together telling a complete story:

1. Anthropic surpasses OpenAI—The power handover in the enterprise market is complete.
2. SpaceX public S-1—Pushing orbital AI computing into the public capital markets.
3. Jensen Huang locks in Vera Rubin capacity—The next round of the chip war has begun.

Collective signal: AI control is shifting from the model layer to the physical layer. The 2024 question was “whose model is smartest,” 2025 is “who can ship products fastest.” After May 28, 2026, the question becomes “who controls the compute infrastructure.”

Musk appears in all three events simultaneously—SpaceX providing compute for Anthropic ($1.25 billion per month), xAI competing with Anthropic, and the SpaceX IPO making him an $800 billion net worth CEO/CTO/Chairman. Is this one person’s compute empire, or a man-made systemic risk? The most noteworthy question for the second half of the year.

📬 This daily report was assisted by AI Daily Digest 3.2.0 Pipeline, manually reviewed. 2026.05.28 Morning Broadcast.