Musk and Anthropic are going to space to find electricity.

Original | Odaily Planet Daily (@OdailyChina)

Elon Musk and Anthropic have actually teamed up!

In the early hours of May 7, Beijing time, a joint announcement from Anthropic and SpaceX shocked the entire AI community. According to the announcement, the two parties have signed a cooperation agreement, and Anthropic will use the full computing power capacity of SpaceX's Colossus1 data center, which will provide Anthropic with over 300 megawatts of new computing power (equivalent to over 220,000 NVIDIA GPUs) within a month, directly improving the user experience for Claude Pro and Claude Max subscribers.

This is the first time Musk's business empire has reached such a direct, formal, and large-scale cooperation with Anthropic. For many years, Anthropic's core partners have been Amazon and Google. Whether in cloud infrastructure, chip supply, or model training, Anthropic has long been deeply bound to AWS and Google TPU systems. Musk has not only never publicly invested in Anthropic but has also criticized its AI safety stance and political tendencies multiple times. There had been virtually no public records of infrastructure cooperation, model cooperation, or business alliances between the two parties until now.

When the most powerful large model company and the most talked-about billionaire in the world finally connect, such cooperation inherently carries a sense of intrigue.

Musk's Delicate Intentions

For Musk, the subtleties of this cooperation lie in the timing.

Just ten days earlier (April 27), Musk's lawsuit against OpenAI officially entered the trial phase in the Northern District Federal Court of California, with the two sides engaging in a flurry of accusations in court in recent days, creating a tense atmosphere. This lawsuit, viewed by outsiders as the "first major case of the AI era," has almost become a comprehensive public display of Musk and OpenAI's long-standing grievances.

And Anthropic, exactly, is currently one of OpenAI's core and direct competitors. Thus, this cooperation naturally takes on an interesting layer of meaning — the enemy of my enemy is my friend; as long as it makes OpenAI uncomfortable, Musk is willing to do anything.

From the perspective of the AI competition landscape, this collaboration also signifies a shift in Musk's strategic layout in the AI era.

On the surface, this is merely a standard computing power transaction — SpaceX provides GPU clusters, and Anthropic gains more inference resources, both parties getting what they need. But it is clearly not that simple.

Because what Musk is doing around AI is no longer just about "model development." Over the past two years, since Musk personally entered the scene with xAI (now renamed SpaceXAI), outsiders have increasingly viewed large model companies like OpenAI and Anthropic as potential competitors to Musk; but with the establishment and use of the supercomputing cluster Colossus, the capabilities of data centers have begun to overflow, and Musk's role has quietly changed.

Musk is increasingly resembling an "arms dealer" in the AI world — anyone lacking computing power can come to him, even if they were once potential competitors.

Anthropic, Deep in the "Dumbing Down" Controversy, Finally Sees a Hero

For Anthropic, the importance of this collaboration may even exceed what outsiders imagine.

In recent months, Claude's reputation has been undergoing subtle changes. On one hand, Claude Opus 4.7 and the mysterious Mythos are still regarded as among the best models on the market; but on the other hand, voices regarding Claude's "dumbing down" have begun to appear more frequently in the community.

Particularly among the heavy developer group, these sentiments are especially pronounced. Some have noticed a "cliff-like drop" in Claude's reasoning ability when handling long code and complex engineering tasks; others have indicated in research reports that the "thinking budget" or response length of some Claude models has been significantly reduced; while more retail users report that Claude's hallucinations are becoming increasingly severe, with the model more frequently providing "seriously nonsensical answers" when processing complex information.

After the "dumbing down" controversy escalated, Anthropic officials released a technical review report, admitting that from March to April, due to product-level adjustments and bugs, Claude models experienced "performance degradation" in complex tasks.

However, this reason did not persuade the market, as public opinion generally believes that the high inference costs and computing power shortages are the primary causes of performance fluctuations in products like Claude in practical applications.

The more powerful the model, the higher the inference cost; the larger the user base, the more terrifying the GPU consumption. All AI companies face a commercial contradiction — users want the model to always operate "at full power," but companies must control costs. Thus, dynamic rate limiting, inference budget adjustments, response length control, priority scheduling... these mechanisms will almost inevitably arise, and the user experience ultimately translates to "the model has dumbed down."

This is why the computing power collaboration with SpaceX is so crucial for Anthropic.

In the announcement, Anthropic stated that with the cooperation agreement in place, the experience of Claude's core users will be directly improved:

First, Anthropic will double the Claude Code usage quota for Pro, Max, Team, and enterprise plans billed by seat from five hours.
Secondly, Anthropic will eliminate throttling measures for Claude Code during peak hours for Pro and Max accounts.
Third, Anthropic will significantly increase the API rate limits for the Claude Opus model.