The Empire of Computing Power vs. The Corporate Assassin: The ultimate showdown between OpenAI and Anthropic for Artificial Superintelligence (ASI) has begun.
A few weeks ago, a landmark event occurred in the AI arena.
Elon Musk, who used to wave his Grok around and claim to challenge OpenAI, suddenly leased his most prominent computing power ace, Colossus 1, to Anthropic.
This is a monster – level cluster equipped with over 220,000 NVIDIA GPUs, which was the iron – heart that supported Musk’s ambition for xAI.
Now, it starts to power Claude.
As recently as February this year, Musk publicly called Anthropic “Misanthropic” (anti – human) and “evil” on X.
Dramatically, his attitude did a 180 – degree turn three months later.
He spent a week having in – depth conversations with Anthropic’s core team and then told the world: “No one has triggered my ‘evil detector’. Claude is probably good.”
Insults are one thing, business is another. More importantly, the enemy of my enemy is my friend.
Musk is suing OpenAI for $150 billion in court and demanding the removal of Sam Altman. The supercomputer he is leasing now belongs to Altman’s biggest rival.
This move kills three birds with one stone: making money, revitalizing idle assets, and strengthening OpenAI’s competitor.
But more importantly, this is a landmark moment for a grander event.
From this moment on, the ultimate showdown in the ASI world has officially converged into a two – horse race: OpenAI, with ChatGPT, Codex, GPT – 5.5, and the Stargate Empire; and Anthropic, with Claude, Claude Code, Mythos, and multi – cloud computing power from AWS, Google, and SpaceX.
On the eve of ASI, the battle between the two companies has already begun.
This article is from the April issue of the New Intelligence Yuan’s ASI Industry Map. In this article, we will focus on the latest progress of ASI and explore in – depth insights into ASI.
The Power Map of Two Super Ecosystems
OpenAI Camp: Microsoft Ecosystem × Stargate Empire
OpenAI has never fought alone. Behind it stands the entire Microsoft Empire, the capital leverage of SoftBank, and a computing power monster under construction – Stargate.
Among them, Microsoft is OpenAI’s oldest ally, having invested over $13 billion in total and deeply embedding GPT into all product lines of Azure, Office, and Copilot.
In the $122 billion super – financing in March this year, Amazon unexpectedly invested $50 billion, NVIDIA invested $30 billion, and SoftBank invested $30 billion. The three giants’ joint bet has pushed OpenAI’s valuation from $500 billion to $852 billion.
In terms of computing power, OpenAI’s Stargate project is an investment marvel: with a total investment commitment of $500 billion, it has signed a five – year, $300 – billion cloud – computing contract with Oracle, and the planned total capacity exceeds 10GW – close to the output of five Hoover Dams. The flagship campus in Abilene, Texas, is already in operation, the GB200 chips are in place, and NVIDIA’s next – generation Vera Rubin platform will start deployment in the second half of 2026.
In addition, OpenAI has formed a “deployment legion” consisting of 19 consulting agencies to help enterprises implement the Codex coding tool, directly competing with Anthropic for the enterprise market.
Anthropic Camp: Mastering Three Clouds × Hedging Across Industries
If OpenAI follows the “empire model” – one sovereignty, one army, one flag, then Anthropic follows the “Swiss bank model”: not exclusively binding with anyone, making everyone feel that investing in it is the safest hedge.
Amazon is its biggest financial backer, having committed over $30 billion in total, with a 5GW Trainium production capacity. The Project Rainier cluster has used 1 million Trainium2 chips to train Claude.
Google has promised a maximum investment of $40 billion, 1 million TPUs, over 1GW of computing power, and a next – generation custom chip in cooperation with Broadcom.
The most interesting part is that Microsoft and NVIDIA ( OpenAI’s core allies ) also promised $5 billion and $10 billion to Anthropic respectively in November last year.
Claude has become the only cutting – edge model in the industry running on three major cloud platforms: AWS Bedrock, Google Vertex AI, and Azure Foundry.
Coupled with the newly acquired Colossus 1 from Musk – Amazon, Google, NVIDIA, Microsoft, and Musk are all betting on Anthropic at the same time.
Venture capital firms and sovereign wealth funds are flocking in: Singapore’s GIC, Coatue, BlackRock, Qatar Investment Authority, Abu Dhabi’s MGX…
In February, it raised $30 billion in Series G, with a valuation of $380 billion. In May, the target for the new round of financing is over $30 billion, and the valuation is heading towards $900 billion – once achieved, it will surpass OpenAI for the first time and become the most valuable AI company in the world.
On May 19th, another bombshell dropped: Andrej Karpathy, a co – founder of OpenAI, officially announced his joining of Anthropic.
The inventor of “Vibe Coding”, an AI educator with 2.5 million followers, and the former head of Tesla’s autopilot, put his educational company on hold and joined Anthropic’s pre – training team to do something very specific: accelerate the training of the next – generation Claude with Claude.
From Jan Leike and John Schulman in 2024 to Andrej Karpathy in 2026, three former core figures of OpenAI have switched to Anthropic one after another – top AI talents are voting with their feet.
Head – to – Head Competition: Who is Winning?
Revenue: Anthropic Achieves the Fastest Growth in Human Business History
Anthropic’s annualized revenue has exceeded $30 billion, and it is expected to reach $10.9 billion in the second quarter, and may even achieve quarterly operating profit for the first time. Four months ago, this figure was $9 billion.
From $9 billion to $30 billion in four months, this is the fastest revenue growth rate in human business history, without a doubt.
In contrast, OpenAI’s annualized revenue is in the range of about $24 – 25 billion – still strong, but its growth rate has been left behind by Anthropic.
ChatGPT has nearly 900 million weekly active users and 50 million paying users, and its dominance in the consumer market is undeniable. However, the problem with the consumer market is that there are many users but a low average customer price.
Anthropic doesn’t have as many users, but it targets the corporate budget – the most lucrative part.
What’s even more thought – provoking is the efficiency gap. In 2025, OpenAI’s training cost was about $20 billion, while Anthropic only spent $5 billion – one – fourth of OpenAI’s cost.
Anthropic has achieved higher revenue than OpenAI with one – fourth of its training cost. OpenAI is expected to break even in 2030, while Anthropic’s goal is 2028 – two years earlier.
Enterprise Market: A Historic Reversal
According to the data from the corporate expense management platform Ramp in May 2026, Anthropic’s enterprise adoption rate is 34.4%, surpassing OpenAI’s 32.3% for the first time. A year ago, the figures were 8% vs 32%, and the gap seemed insurmountable.
What’s even more fatal is that when enterprises choose an AI provider for the first time, 70% choose Anthropic.
A report from Menlo Ventures shows that in the enterprise coding market, Claude’s share is between 42% – 54%, while OpenAI only has 21%. The annualized revenue of Claude Code alone has reached $2.5 billion.
Ironically, Microsoft – OpenAI’s biggest ally – cancelled the internal engineers’ usage licenses for Claude Code due to the exploding token costs and required them to switch to its own tools. Even Uber couldn’t hold on: its annual AI budget was exhausted in April, and heavy users spend $500 – $2000 on Claude Code per month.
But this just proves one thing: Programmers can’t do without it.
Models and Products: Full – Stack Behemoth vs. Corporate Assassin
At the model level, the two companies have played completely different hands.
OpenAI’s ace is GPT – 5.5, positioned as a full – stack model for software engineering, scientific research, and complex professional work.
Combined with Codex (with over 4 million weekly users) and a hybrid – cloud solution launched in cooperation with Dell, OpenAI aims to create an all – in – one AI package that can do everything.
Anthropic takes a completely different route. Claude Opus 4.7 focuses on autonomous, long – term agent work and in – depth knowledge processing; Claude Code has achieved an industry – leading 80.8% on the SWE – bench; Mythos directly targets network security – it discovered over 10,000 high – risk vulnerabilities in the “Project Glasswing” plan, which prompted the European Central Bank to rarely convene banks to accelerate patching, and the Financial Stability Board to request a briefing.
The newly exposed Claude Memory Files – a file – based memory system similar to a personal Wiki – is laying the foundation for the next – generation long – running AI agents.
In a nutshell: OpenAI is building a super – application, while Anthropic is building a corporate operating system. One wants the whole world to open ChatGPT every day, and the other wants every company to rely on Claude every day.
When Will ASI Arrive?
The ultimate mystery of all this competition is: When will Artificial Superintelligence (ASI) arrive?
The data from the AI safety assessment agency METR provides the most intuitive reference: the “time horizon” for AI to autonomously complete programming tasks (i.e., how long an AI can independently complete tasks that human experts do) is growing exponentially –
- 2019 (GPT – 2 era): 3 seconds
- End of 2025 (Claude Opus 4.5): Nearly 5 hours
- Doubling speed: Accelerated from doubling every 7 months to every 4 months
- Claude Mythos Preview: Already approaching 16 hours
In other words, the “independent working time” of AI is not increasing linearly but is exponentially breaking through the time boundaries of human tasks. It’s 5 hours today, and tomorrow it could be a day, a week, or a month.
该文观点仅代表作者本人，36氪平台仅提供信息存储空间服务。
