Article | MostTalk, Author | He Yiran, Editor | Liu Yuxiang
Now, Anthropic is not only a favorite among programmers but also a darling of tech titans.
Because it helps them boost their stock prices.
On May 6 local time, Musk announced the formal dissolution of his AI company xAI as an independent entity and its full integration into SpaceX, becoming an AI product under SpaceX named SpaceXAI. Simultaneously, SpaceXAI announced a global top-tier computing partnership with Anthropic: the full computational power of the Colossus 1 supercomputing cluster will be exclusively leased to Anthropic for Claude series model inference services.
Located in Memphis, Tennessee, Colossus is one of the world's largest and fastest-deployed artificial intelligence supercomputing clusters. The Colossus 1 cluster is equipped with over 220,000 NVIDIA GPUs, including densely deployed H100, H200, and next-generation GB200 accelerators.
In return, Anthropic indicated it has "expressed interest in collaborating with SpaceX to develop multiple gigawatts of orbital AI computing power."
Not only Musk, but Google has also long supported Anthropic. On April 25, Google further doubled down, planning to invest up to $400 billion in Anthropic and provide 5 gigawatts of computing support. Google pledged an immediate $100 billion cash injection at a $350 billion valuation, consistent with the valuation from the February financing round. Upon Anthropic meeting its performance targets, Google will invest an additional $300 billion while significantly expanding its computing scale.
Although Google's own Gemini model ranks among the so-called "Big Three," it still heavily invests in Anthropic. According to public information, Google's parent company Alphabet holds approximately 14% of Anthropic's shares. Amazon holds about 15%–16%, making it the largest shareholder, with cumulative investments exceeding $80 billion, and provides computing power and cloud service support.
It's easy to understand Musk's business move of dismantling xAI just before SpaceX's impending IPO. xAI has already fallen out of the AI top tier, with a peak valuation of no more than $250 billion. It makes more sense to lease the Colossus 1 supercomputing cluster to Anthropic in exchange for Anthropic's orders for SpaceX's future space-based computing power, keeping SpaceX's commercial story alive.
After all, early investors in Anthropic like Google and Amazon have reaped enormous returns. Alphabet's first-quarter profits soared 81% to $62.6 billion, while Amazon's net profit surged 77% to $30.3 billion. Of these profits, $28.7 billion and $16.8 billion respectively stemmed from adjustments in the equity value of Anthropic.
Beyond the paper gains from Anthropic's valuation, there are also real cash orders. Anthropic has committed to paying Google $200 billion over the next five years to use its cloud servers and chips. Anthropic's deal accounts for over 40% of the future revenue commitments disclosed in Google's latest earnings call. Amazon Web Services also has a backlog of computing demand orders from Anthropic.
The stock prices of Google, Amazon, and Tesla (Musk), who have heavily invested in Anthropic, have shown strong recent performance. However, the stock of Microsoft, an ally of OpenAI, has been relatively sluggish, despite it having also invested in Anthropic during the G round, becoming a minor shareholder.
The AI "Cold War" may have already erupted, with the two camps being Anthropic-Google-Musk, abbreviated as Camp A, and OpenAI-Microsoft-SoftBank, abbreviated as Camp O. Interestingly, cracks have already appeared within the OpenAI camp.
This is the time for Camp A's major offensive.
01
On April 28 local time, Musk's lawsuit against OpenAI officially went to trial at the Oakland Federal Court in California.
Musk testified for three consecutive days, accusing how he invested huge sums based on his belief in OpenAI's non-profit promise. He demanded the court order the removal of Sam Altman and Greg Brockman from management, restore OpenAI as a purely non-profit research institution, recover and return the technological and commercial value corresponding to his $38 million donation to charity, and most importantly, force OpenAI to open-source and disclose all AI technical details and training data sources.
This lawsuit is far more than a personal feud between Musk and Altman; it is a precise attack on OpenAI. The timing, deliberately set during OpenAI's sprint towards an IPO, clearly aims to disrupt the commercialization process of this AI industry leader and lay the groundwork for the Anthropic-Musk alliance.
American courts do not allow photography, so media had no visuals from the courtroom. Yet, social media was flooded with various convincingly fake photos of Musk and Altman.
These highly realistic images were all generated by OpenAI's unexpectedly launched ChatGPT Images2.0 in April. This thinking-capable image model achieves "unprecedented specificity and fidelity." As Altman put it, it's like "a leap from cave paintings to the Renaissance."
Although Images2.0 stunned social circles, the capital market didn't show sufficient excitement. The reason is simple: Images2.0 has limited quotas. Free users can generate only 3 images before needing to upgrade to the paid ChatGPT Plus. For users chasing novelty, grabbing the freebie is enough, making it increasingly difficult for OpenAI to extract substantial commercial returns from its vast user base.
Looking at the large model traffic share, by March 2026, ChatGPT's share had dropped to about 56.7%, while Google's Gemini climbed to 25.5% and Anthropic's Claude rose to 6%. ChatGPT's once-overwhelming dominance has clearly diluted.
Even more concerning is OpenAI's revenue-generating ability. Its perennial loss-making situation is gradually wearing out investor patience. Third-party statistics show that in Q1 2026, Anthropic surpassed OpenAI with a 31.4% share of global large model market revenue versus OpenAI's 29%.
It's important to note that Anthropic's monthly active users are only 134 million, with a monthly average revenue per user (ARPU) of $16.20, while OpenAI's is only $2.20. The difference in the two companies' revenue performance is not merely a question of whether B2C or B2B is superior, but whether they can make customers accept their business model through their products. Anthropic focuses on high-certainty paying scenarios and high-willingness-to-pay user groups, establishing a commercial flywheel.
Another factor affecting revenue is the joint announcement by Microsoft and OpenAI in April to formally terminate their seven-year exclusive binding model. Microsoft will no longer pay a share of OpenAI's revenue, while OpenAI's revenue sharing to Microsoft will continue until 2030.
To improve revenue, while updating GPT-5.5, OpenAI also made updates to its advertising business. The ad platform ended its testing phase and fully opened to U.S. businesses. Advertisers can self-register, set budgets and bidding strategies, and purchase ad space on a cost-per-click (CPC) or cost-per-thousand-impressions (CPM) basis.
A clear "dual-track system" will be established for AI answer streams. Ads will appear as sponsored modules after AI answers, intelligently displayed based on conversation context. OpenAI emphasized that the ad system is independent of answer generation; advertisers cannot influence AI output nor see user personal data.
If 2023 was the AI industry's moment of glory, in 2026 everyone is doing the math.
Over the past few years, any startup touching the "AI" concept could attract capital frenzy, and internet giants could expand their imagination limitlessly. But by 2026, myths are gradually shattering. Most AI unicorns have become "bubbles," and the balloons inflated by tech giants are being pulled back to earth.
The rules of the growth-at-all-costs era are gradually coming to an end. The core issue facing the industry is not who can build more stunning models, but who can make enterprise customers willingly pay.
Especially after DeepSeek-V4 was released and open-sourced, it established a冷酷的 "kill line": The commercial value of any closed-source model with performance inferior to DeepSeek V4 will be zero.
This directly and indirectly accelerated the consolidation of various forces towards the top players Anthropic and OpenAI, also forcing OpenAI to accelerate its commercialization efforts.
02
Those who haven't decided which side to join are perhaps the most anxious.
The most typical example might be Meta.
According to Q1 earnings data, Meta's total revenue grew approximately 33% year-over-year. While this number seems decent, it conceals a significant crisis — daily active users (DAU) decreased by about 20 million sequentially, marking the first quarterly sequential decline since this metric was publicly disclosed. Meta's AI investment return path relies entirely on improved ad conversion rates and increased social platform user engagement. A decline in DAU is an extremely dangerous signal.
Meta stated that its full-year capital expenditure expectation was raised from approximately $115-135 billion to $125-145 billion. The increase is mainly due to rising component prices and additional data center costs. This implies Meta's costs are not actively expanding strategic advantages but passively absorbing supply chain price hikes.
A more critical blow came from China's halt of Meta's planned $2 billion acquisition of Manus. According to regulatory requirements, Meta must cancel the acquisition deal, refund the paid acquisition price, delete related products and data, and return Manus's technology ownership.
At the critical turning point where the AI industry shifts from "storytelling" to "accounting," unquantifiable uncertainty becomes the weakest link most intolerable to the capital market. After the earnings release, Meta's stock price directly fell about 7%, nearly wiping out all gains since the beginning of the year.
In contrast, Google's Q1 earnings were more favored by the capital market.
Although the progress of the Gemini model is currently the slowest among the "Big Three," fortunately, Google's ecosystem and scenarios are so rich that it has deeply embedded the Gemini model into its search, cloud, and subscription systems, making it an amplifier for existing business growth — paid monthly active users for Gemini Enterprise grew about 40% sequentially; enterprise AI solutions are becoming a major driver for cloud business; and search ads remain strong driven by AI.
More importantly, it invested in Anthropic and holds Anthropic's hundred-billion-dollar order, which the market rewarded positively.
Amazon, leveraging its cloud business advantage, has taken a risk-hedging route in the AI infrastructure era. While investing in OpenAI, making it AWS's exclusive third-party cloud distribution partner, Amazon continues to secure its hundred-billion-dollar computing agreement with Anthropic. In Andy Jassy's view, he doesn't want to bet on which model will be the ultimate winner, but ensures different winners all need to run through AWS.
The value of AI must ultimately be validated by commercial implementation. The AGI story envisioned by OpenAI, with global monopoly as its underlying logic, has largely been proven to fail in practice. AI infrastructure construction and compliance issues need to be handled in a distributed manner across regions and fields within a global layout.
Anthropic's rise to prominence is clearly reshaping Silicon Valley's power structure. Over the past year, Anthropic's valuation has increased tenfold, reaching a peak of $1.2 trillion, making it the hottest super-unicorn. Its Q1 2026 revenue surged 80-fold year-over-year, far exceeding the original 10x target, with full-year 2026 revenue projected at $18 billion.
2023 saw OpenAI's solo sprint, 2024 was a global free-for-all, 2025 had the "Big Three," and this year shows an emerging duopoly.
It's very possible, and not impossible, that next year Anthropic will lead the race.
03
The anchor point of the AI narrative is shifting from NVIDIA to Anthropic.
Over the past two years, NVIDIA's GPUs were the starting point of all large model stories — whoever had more computing power had the say. But now, computing power itself is becoming commoditized. What is truly scarce is no longer the chips, but the application-layer leading companies that can transform computing power into sustained revenue. Anthropic is precisely the bearer of this new narrative center.
The most subtle change is happening on the capital side. Currently, Google, Amazon, and SpaceX have all positioned "investment in Anthropic" as a core story in their earnings reports for investment returns and computing orders. This deep binding goes beyond ordinary financial investment — changes in Anthropic's valuation directly impact the quarterly profits and future expectations of these three giants.
Data tells the story. Over half of Alphabet's Q1 net profit came from equity investment paper gains, with Anthropic and SpaceX contributing the largest share. Amazon's Q1 net profit surged 77% to $30.255 billion, with a pre-tax gain of $16.8 billion originating from the valuation adjustment of Anthropic. Although SpaceX is not yet public, in its upcoming IPO roadshow, the annual $3-4 billion in computing lease revenue from Anthropic and promised space computing orders have become key narratives supporting its $1.75-2 trillion valuation.
Once this structure solidifies, Anthropic is no longer merely an AI company, but a "systemic financial node." Any negative news about it — whether technological bottlenecks, security incidents, or loss of core talent — will no longer be its own affair. Instead, it will simultaneously impact the tech sector through three transmission paths: the book value of Google's equity investment, the fulfillment expectations of Amazon AWS's computing orders, and the IPO pricing logic of SpaceX.
In contrast, the relative stock price weakness of Microsoft, an ally in the OpenAI camp, is precisely due to the lack of this "equity-orders-profits"联动闭环 (linkage closed loop). Microsoft and OpenAI terminated their exclusive binding in April, with Microsoft no longer sharing OpenAI's revenue, only retaining OpenAI's ongoing revenue share obligation to Microsoft. No matter how high OpenAI's valuation climbs, Microsoft's balance sheet won't reflect corresponding investment gains.
When the entire market's capital constructs a positive cycle around Anthropic, Microsoft's position appears exceptionally passive.
However, when the entire market bets on the same winner, risk is also amplified simultaneously.
Anthropic's valuation premise — maintaining technological leadership and revenue growth — is proven by its computing orders and equity commitments to the three giants. This is a self-reinforcing cycle, but also a self-fragilizing structure. Any loosening in one link, such as the next-generation Claude underperforming expectations, or open-source models achieving performance parity with Anthropic's, could trigger a chain reaction — from Anthropic's own valuation correction to the earnings and stock price performance of the three giants.
The AI industry is moving from "multi-party free-for-all" to "one superpower, multiple strong players," but the capital market doesn't seem ready to accept one fact: when all eggs are placed in one basket, that basket itself becomes the biggest risk.
DeepSeek's open-source "kill line" hangs like a Sword of Damocles over Anthropic. As long as the open-source community can maintain performance close to or on par with it, Anthropic's commercial value "anchor" could loosen at any time.
Of course, Anthropic is well aware of this, as are its allies and "hidden allies" in the Trump administration. Therefore, continuing to suppress open-source models like DeepSeek and China's domestic computing and storage will be their long-term strategy.
This is not just about the U.S. stock market's AI narrative; it's a war over technological hegemony.
