SpaceX's IPO pricing countdown phase has seen two large, month-by-month compute contracts disclosed via U.S. Securities and Exchange Commission (SEC) filings in quick succession. The first, revealed in the May 20 S-1 filing, is a contract with Anthropic for $1.25 billion per month, leasing all compute capacity from xAI’s Colossus 1 data center located in Memphis, Tennessee. The second, disclosed in the June 5 S-1/A amendment, is a Google contract worth $920 million per month, covering approximately 110,000 Nvidia GPUs.
Combined, these two contracts amount to $2.17 billion monthly, or $26 billion annually. If not terminated early, the total value over a three-year term exceeds $70 billion. In its S-1 filing, SpaceX’s official description of the Anthropic contract emphasizes “monetizing idle infrastructure capacity,” placing particular emphasis on the word “idle.”
SpaceX’s xAI division reported an operating loss of $2.47 billion in Q1 2026. According to data disclosed by SpaceX in the S-1, the department spent $12.7 billion on AI capital expenditures in 2025 and another $7.7 billion in Q1 2026. BitMEX’s analysis of the S-1 shows that SpaceX’s cumulative deficit had reached $41.3 billion by the time of pricing.
The turning point occurred on April 1. On that day, SpaceX confidentially filed its IPO registration statement with the SEC. On May 20, the S-1 was publicly disclosed, along with the announcement of the Anthropic contract; on June 1, the S-1/A amendment was submitted; on June 3, the share price was set at $135 per share; on June 4, roadshow activities commenced; on June 5, the Google contract was disclosed; on June 11, pricing was finalized; and on June 12, Nasdaq debut trading began under the ticker symbol SPCX.
Indian financial platform IndMoney’s interpretation of this timeline cuts to the core: “Three weeks ago, xAI appeared to be one of SpaceX’s heaviest financial burdens. Now, it has approximately $2.17 billion in monthly compute revenue from two creditworthy clients. This isn’t cosmetic storytelling—it’s structural rewriting.”
The disclosure window for both contracts is highly concentrated: the Anthropic contract was made public 22 days before IPO pricing, while the Google contract was disclosed just six days prior.
The foundational asset underpinning both contracts is a facility that Musk himself has already vacated.
Colossus 1 is located in Memphis, Tennessee. Built by xAI in just 122 days by December 2024, it houses over 220,000 Nvidia GPUs (a mix of H100, H200, and GB200 models), with a power capacity of 300 megawatts. Originally, this facility served as the core compute infrastructure for training xAI’s proprietary large language model, Grok.
However, prior to signing the contract with Anthropic, Colossus 1’s utilization rate had dropped to around 11%. DataCenterDynamics cited Musk’s own statement: “After that, leasing Colossus 1 to Anthropic was acceptable to me because SpaceXAI had already migrated its training workloads to Colossus 2.”
xAI has since expanded its data center footprint around Memphis. Colossus 2 went live in January 2026, and a third data center is being developed in Southaven. The company also acquired land adjacent to the site for $659 million to construct another building. Grok, the original product served by Colossus 1, has seen declining usage. According to TechCrunch, usage of xAI’s flagship AI assistant Grok has dropped sharply over recent months, freeing up server capacity—now being sold to one of its closest competitors.
The Google contract, disclosed later but closer to the IPO window, contains several structural nuances worth unpacking.
According to SpaceX’s June 5 SEC filing, the contract is valued at $920 million per month, covering the period from October 2026 to June 2029. SpaceX must complete delivery by September 2026; otherwise, Google may terminate the contract or accept fewer GPUs. Starting in 2027, either party may terminate with 90 days’ notice. The leased asset comprises approximately 110,000 Nvidia GPUs.
Google’s official statement to The Wall Street Journal described the arrangement as “a short-term, timely solution to ensure we have bridging capacity to meet the surge in demand for our Gemini Enterprise agent platform—demand that has exceeded our expectations.”
The choice of terms like “short-term” and “bridging capacity” contrasts with SpaceX’s own characterization of the contract as part of “contractualized monthly committed revenue” in its prospectus. This reveals a divergence in how the parties perceive the nature of the agreement.
The contractual background also involves equity stakes: Google is an early investor in SpaceX, holding approximately 5% of shares, and Google executive Donald Harrison serves on SpaceX’s board. This means Google has direct financial interest in SpaceX’s IPO pricing.
The Anthropic contract is the larger of the two, disclosed earlier, and most closely scrutinized.
Per SpaceX’s S-1, the contract is valued at $1.25 billion per month through May 2029, with discounted pricing for the first two months. Either party may terminate with 90 days’ notice. The total contract value exceeds $40 billion. French AI media ActuIA estimates an implied cost of approximately $7.78/GPU/hour.
Anthropic is leasing the entire compute capacity of Colossus 1 for inference workloads, not model training. According to Basenor’s reporting, this compute will be used to expand usage quotas for Claude Pro and Claude Max subscription tiers. This distinction is critical: using a competitor’s compute for inference (responding to user queries) represents a fundamentally different dependency than leasing compute for training proprietary models.
Comparative scale provides context: publicly reported, CoreWeave’s compute contract with OpenAI is valued at around $11.9 billion over five years. The Anthropic-xAI/SpaceX contract is approximately 6.3 times larger in scale.
A deeper layer of context lies in Musk’s prior public remarks about Anthropic. Earlier this year, Musk repeatedly labeled Anthropic as “evil” across multiple public appearances. From public condemnation to signing the largest single AI compute contract in history, the shift occurred within just a few months.
The shared structural feature of both contracts—the ability for either party to terminate with 90 days’ notice—creates a notable contrast with SpaceX’s IPO valuation narrative.
The standard financing logic for data centers relies on stable cash flows from long-term anchor customers, typically locked in 10+ year contracts aligned with power, construction, and depreciation cycles. Both the Anthropic and Google contracts deviate significantly from this norm.
Each party has reserved “soft” exit justifications: Anthropic cites inference workloads (which are elastic and subject to fluctuating subscription demand); Google characterizes the arrangement as “short-term bridging capacity.” Yet SpaceX’s S-1 presents this as “contractualized monthly committed revenue” to IPO investors, portraying a predictable, recurring income stream.
SpaceX’s IPO pricing itself rests on several forward-looking assumptions. According to CNBC’s report on June 3, the $1.77 trillion valuation “assumes completion of EchoStar spectrum and Cursor-related transactions.” Morningstar’s research concludes that SpaceX is “overvalued” and advises investors to wait until after the IPO before considering purchase. Motley Fool analyst Adam Spatacco recommends “initial caution,” citing that the company is issuing only about 4% of its public float, with significant additional unlock events for early investors and employees expected within the first six months post-IPO.
In underwriting, Goldman Sachs leads with Morgan Stanley, Bank of America, Citigroup, JPMorgan Chase, and 17 other banks participating. Retail investor allocation reaches 30%, three times the typical level for mega-cap IPOs (~10%). SpaceX completed a 5-for-1 stock split on May 4.
Author: Ada, DeepTide TechFlow
Disclaimer: Contains third-party opinions, does not constitute financial advice
Pharos Network Unlocks AI Model Payment Channels, Introduces New Use Cases for $PROS and USDC as Platform Payment Instruments
11 days ago
NVIDIA Has Plenty of Cash—Why Is It Borrowing $20 Billion?
11 days ago
Will Claude Ban Accounts and Verify ID Cards? Face Recognition Was Old News from Two Months Ago, and "Handing Over Data to Police" Is a Misinterpretation
11 days ago
Japan's Central Bank on the Brink of Rate Hike—Can the AI Bull Run Withstand?
11 days ago
5-Second Breakthrough with Just 1 Interaction: Has the "Strongest Security Mechanism" of Claude Fable 5 Been Cracked by a Chinese Team?
13 days ago
Why Is the "AI Service Subscription Model" Inevitably Headed for Extinction?
13 days ago
Managing a company valued at nearly a trillion dollars, Anthropic's CEO has only one direct report.
13 days ago






