Google to pay SpaceX $920M a month for compute capacity at xAI data centers
Google’s plan to pay SpaceX about $920 million a month for access to 110,000 Nvidia GPUs has raised questions over whether this is genuinely about meeting AI compute demand or primarily a way to boost SpaceX’s revenue and valuation ahead of its IPO, in which Google is a major shareholder. Commenters debate if this constitutes circular or conflicted financing, note that xAI’s Grok appears to have flopped and left large data centers underused, and highlight wider concerns about an AI-driven GPU bubble, index-fund exposure, and the environmental impact of gas-powered data centers.
Deal structure and pricing
- Google will pay SpaceX about $920M/month (from late 2026 to mid‑2029) for access to ~110k Nvidia GPUs and associated infrastructure; there is a 90‑day termination option for both parties after 2026.
- Commenters calculate this as roughly $11–12 per GPU‑hour—seen as very high but possibly justified by severe supply constraints and “bridge” capacity needs.
- Some note SpaceX’s power setup (trailer‑mounted gas turbines) and datacenter build‑out likely make their costs unusually high; others say regular GPU rentals are already below cost elsewhere.
Circular financing, valuation, and IPO optics
- Many see this as valuation engineering: Google owns ~5–7% of SpaceX, so boosting SpaceX revenue by ~$11B/year at a ~90x sales multiple could add tens of billions to Google’s stake on paper.
- Others argue this is not “circular financing” in the strict sense: it’s an affiliate transaction with potential conflicts, but cash only flows Google → SpaceX, not back.
- There is heavy debate over whether the deal mainly exists to push SpaceX into 12 months of GAAP profitability and fast‑track index inclusion at an “insane” multiple, versus being a straightforward response to real compute demand.
xAI/Grok and excess capacity
- A recurring theme: xAI overbuilt GPU capacity for Grok, which has relatively low adoption, leaving large idle clusters (Colossus 1/2) now being leased to Anthropic and Google.
- Some view this as a rational pivot to a CoreWeave‑style “neocloud”; others see it as evidence xAI failed as a frontier lab and is now a commoditized host.
Compute scarcity and Google’s motives
- Several commenters think this simply reflects extreme GPU, memory, power, and permitting constraints; even hyperscalers are short on datacenter capacity.
- The contract’s short cancellation window is cited as evidence that Google sees it as temporary “bridge” compute and optionality, possibly also to deny capacity to competitors.
- Some speculate Google might resell capacity via GCP; others think it’s needed to shore up Gemini and agent platforms amid reliability and throttling issues.
Ethics, environment, and politics
- Strong criticism of Google’s “carbon‑free by 2030” branding while renting from gas‑fired datacenters, including ones alleged to be operating illegally and near low‑income neighborhoods.
- Broader discomfort with funding Musk‑linked ventures (described by some as grifts, financial engineering, or “cult stocks”), especially given index‑fund exposure and perceived regulatory capture.
Orbital datacenters debate
- SpaceX’s pitch of future space‑based datacenters is widely discussed; many call it physically or economically implausible (cooling, radiation, launch costs, maintenance).
- A minority argue it could become viable if launch costs and radiator tech improve and terrestrial siting becomes politically untenable; most remain skeptical.