Version 2

SpaceX's S-1 IPO filing, published May 20, 2026, confirmed a $1.25 billion-per-month Cloud Services Agreement with Anthropic for COLOSSUS and COLOSSUS II compute capacity through May 2029 — roughly $45 billion total — but a CNBC report from May 7 suggests the deal was publicly announced weeks before the S-1 dropped, and also includes a space development component [1][2][3]. The filing further revealed SpaceX absorbed xAI into a new SpaceXAI division and reframed its launch and satellite businesses as supporting infrastructure for AI [11]. xAI burned $6.4 billion last year [15], Morningstar called the combined entity's finances 'reckless' [16], and the IPO is targeting a valuation of $1.75–2 trillion [4][14] — even as Reuters reports Anthropic itself is nearing its first quarterly profit [21]. The SpaceX deal is one leg of Anthropic's multi-provider compute strategy that also includes a $33 billion Amazon commitment [18] and a large Google Cloud arrangement [19].

The story has expanded from a single deal into a window on how much frontier AI infrastructure actually costs: Anthropic alone has locked in over $90 billion in contracted compute across SpaceX, Amazon, and Google, revealing a capital commitment race with no near-term ceiling. The financial paradox is now fully visible — SpaceX's model business (xAI/Grok) is losing billions while its infrastructure arm collects billions from the AI labs it competes against, and investors are being asked to price that contradiction at up to $2 trillion.

The Anthropic–SpaceX compute deal entered the public record in two phases. A CNBC report published May 7, 2026 described a joint announcement from Anthropic and SpaceX covering a compute partnership that includes a space development component — details not present in earlier coverage [1]. The full financial terms became visible on May 20 when SpaceX published its S-1 IPO filing: Anthropic agreed to pay $1.25 billion per month for capacity on the COLOSSUS and COLOSSUS II data centers in Memphis, Tennessee, running from May 2026 through May 2029 and totaling approximately $45 billion [2][3]. Capacity is ramping at reduced fees through mid-2026 before reaching full monthly payments, and either party can exit with 90 days' notice [3]. Colossus 1's full capacity is dedicated to Anthropic [4][5], while COLOSSUS II is equipped with NVIDIA GB200 NVL72 systems also used to train Grok 5 [6]. xAI itself published a statement about the compute partnership [7], and the Grok chatbot was widely observed publicly confirming deal terms in social media replies [8][9][10] — an unusual form of first-party disclosure.

The S-1 simultaneously disclosed that SpaceX has formally acquired xAI and integrated it into a new SpaceXAI division overseeing the Grok model family [11]. The filing's strategy section is titled 'Our AI Compute Infrastructure Advantage and Growth Strategy' and positions SpaceX's launch and Starlink businesses as supporting infrastructure for its AI ambitions rather than the core business [12][11]. The company claims a $26.5 trillion total addressable market and cites SemiAnalysis research showing data center construction accounts for roughly 30% of AI compute capex for most companies lacking SpaceX's build-cost advantage [13]. Valuation targets referenced across financial media range from $1.75 trillion [4] to $2 trillion [14], which would make it among the largest IPOs in history.

The financial disclosures in the S-1 have generated a second wave of skeptical analysis that goes beyond Ars Technica's initial model-quality critique. xAI burned $6.4 billion last year and the filing signals that spend trajectory is not declining [15]. Morningstar responded with a report headlined 'Financials Look Reckless: Lifting xAI's Hood in the SpaceX IPO' [16], adding institutional credibility to concerns about whether infrastructure revenue can justify the losses in the model business. Competing reporting offers a partial counterpoint: a LinkedIn analysis argues that Grok has been gaining ground on ChatGPT and Gemini, though the headline notes the path 'isn't pretty' [17], implicitly acknowledging the market share gains may be driven by bundling with X rather than organic product adoption.

Zooming out, the SpaceX deal is one piece of a far larger compute acquisition strategy by Anthropic. The company has a reported $33 billion commitment with Amazon [18], a substantial Google Cloud arrangement [19], and the $45 billion SpaceX contract — a total contracted compute footprint exceeding $90 billion across three providers [20]. Against this backdrop, Reuters reported that Anthropic is approaching its first quarterly profit [21], suggesting its revenue growth may be outpacing even these extraordinary infrastructure costs. That profitability milestone, if confirmed, would reframe the entire story: a company spending at this scale and still approaching profit implies either extraordinary pricing power on its AI products or a revenue ramp that has surprised even internal projections.