SpaceX is bringing its hugeComputing power infrastructureinto a high-speed cash machine. On the eve of landing in the U.S. stock market, Musk's SpaceX has signed huge amounts with Anthropic and Google.computing power leasingThe agreement, with a total value of more than $70 billion and an annualized revenue of $26 billion, injects strong fuel into its IPO narrative.

On June 6, according to Reuters, SpaceX announced on June 5 that it had signed a multi-year cloud service agreement with Alphabet's Google. According to the regulatory filing, Google will pay SpaceX $0.92 billion per month from October of this year through June 2029 for about 110000 blocks.NVIDIAGPU and related computing resources.

Previously, Anthropic had announced in May that it would lease SpaceX's location in Memphis, Tennessee, for $1.25 billion a month.Colossus 1 Data CenterThe initial commitment period is six months.Combined, the two agreements would give SpaceX about $2.17 billion per month in revenue from its computing leasing business, with an annualized size of about $26 billion; if both contracts were executed to maturity, the total contract value would exceed $70 billion.

The above transaction directly reinforces the AI growth story of SpaceX's sprint IPO. SpaceX plans to complete the listing next week, with a target financing scale of $75 billion million, which is expected to be the largest IPO in the history of US stocks.The explosive growth of the computing power leasing business has provided it with quantifiable high-visibility revenue, but at the same time, doubts about SpaceX's AI laboratory xAI's own research and development capabilities are quietly heating up.

Google Big Order: $0.92 billion/Month, Locked to 2029

According to SpaceX's filing with regulators, the Google Cloud Services Agreement was formally signed on June 5, 2026. The agreement stipulates that Google will use computing power at a lower rate before September this year and enter the climbing stage; from October to June 2029, it will pay a fixed monthly payment of $0.92 billion.

The computing resources provided include about 110000 NVIDIA GPU, CPU, memory and other related components.

The terms of the agreement also have protective clauses: if SpaceX fails to provide the agreed number of GPU before September 30, Google can immediately terminate the agreement after the expiration of the one-month grace period, or choose to accept the actual number of GPU provided and reduce the monthly fee proportionally.

In addition, after December 31, 2026, either party may terminate the agreement with 90 days' notice to the other party. Google will retain ownership and all intellectual property rights in its content, AI models and associated data.

According to the analysis, this means that the total value of the $70 billion contract disclosed by SpaceX depends to a certain extent on whether the two parties choose to fulfill the agreement to the expiration date.

The Anthropic-SpaceX computing agreement was announced in May that Anthropic would lease most of the computing power at SpaceX's Colossus 1 facility for $1.25 billion per month.

The facility, located in Memphis, Tennessee, has more than 220000 NVIDIA processors and can provide 300 megawatts of new computing power for Claude chatbots, which will be delivered within a month.

Anthropic big single behind another secret

According to the technology media The Information, there is a little-known competitive entanglement behind the deal. For a long time, xAI has tracked and tried to use Anthropic technology to develop its products-including the continuous use of its model output in multiple ways even after the Anthropic has cut off access.

According to reports, six people familiar with the matter disclosed that xAI had carried out a months-long distillation project and directly used Claude's response results to train its own programming model. In January, xAI co-founder Tony Wu announced to employees that Anthropic had cut off access, but some xAI engineers then continued to access Claude through personal accounts until they were closed in Anthropic. According to two people familiar with the matter, xAI also used middlemen.Blackbox AIAccess to Anthropic model, the last use record can be traced back to the middle of May this year.

Musk himself publicly admitted in a court deposition in May that xAI had been "partially" trained with OpenAI models.GrokIt is said that such practices are widespread in the industry. However, neither he nor the company disclosed the specific extent of xAI's use of Anthropic technology or the various workaround methods used by employees to gain direct access.

xAI hardware planning error, spawning SpaceX computing power leasing business

The Information report reveals that xAI is facing a series of serious challenges at the AI R & D level, which is in tension with the AI growth story SpaceX told Wall Street.

InPersonnel level, xAI experienced massive upheavals:A large number of core staff, including eight co-founders, have left, and the engineering team has been cut.

According to reports, a person familiar with the matter disclosed that as of mid-May this year, the number of xAI pre-training teams was less than five, compared with more than 20 a year ago.Grok CodeProject leaders have been replaced at least four times in a few months, with the most recent one, Beibin Li, leaving in May. In addition, an employee accidentally deleted core training data for the AI programming product during the data migration process, resulting in a loss equivalent to two to three weeks of work.

InStrategic levelMusk said at an internal meeting earlier this year that he planned to introduceCursorAnd Mistral AI fill critical technical gaps-Mistral is responsible for pre-training and Cursor is responsible for post-training. However, early contacts with Mistral did not translate into formal cooperation: Mistral executives visited xAI headquarters in mid-April, but high-level talks between the two sides were interrupted several weeks later. Mistral co-founder engineer Devendra Chaplot joined xAI only about a month before leaving in May.

The cooperation with Cursor has deepened.In April, SpaceX announced that it would open up its infrastructure to Cursor to train its own model Composer and was given the option to acquire Cursor for $60 billion later this year, with a $10 billion severance fee if it was abandoned. Cursor CEO Michael Truell is now a regular guest at the xAI office.

According to The Information, xAI has suffered a series of recent setbacks,It has raised questions about whether SpaceX will focus on its own AI research and development or its computing power leasing business.However, Anthropic's payment to SpaceX went some way to making up for the financial shortfall in xAI, a deeply money-losing division that had led to a big loss for the overall company in the first quarter.

However,SpaceX will calculate the power behind the external rental, reflecting the xAI's own technical difficulties.According to reports, sources have revealed that xAI encountered severe obstacles while training the Grok model at the Colossus 1 data center—The facility employs a heterogeneous mix of GPUs based on the H100, H200, and GB200 architectures, and this heterogeneous environment has made it difficult to advance training effectively, prompting xAI to migrate its model-training workloads to the Colossus 2 data center.

It was precisely this design and planning flaw that provided SpaceX with an opportunity to advance its commercial strategy.SpaceX is attempting to turn xAI’s idle computing assets into a revenue stream by signing cloud service agreements with several AI-related companies,Monetize the idle computing power of Colossus 1.

However,SpaceX’s upcoming IPO highlights compute‑power leasing revenue as a key growth driver, but the rapid expansion of this business model has also raised questions about its strategic positioning in the AI sector.

On the X platform, Musk has repeatedly described the Anthropic agreement as a “short-term arrangement,” adding, “I think we may at some point need to reclaim the computing power.” SpaceX also stated in its IPO filing,The company is continuously training the next-generation version of its flagship Grok model, but has not disclosed any release timeline.