Google to Pay SpaceX Nearly $1 Billion Monthly in Cloud-Computing Deal

On June 5, 2026, a filing buried in SpaceX's amended S-1 registration statement revealed a contract that dwarfs anything previously seen in cloud computing: Google will pay SpaceX roughly $920 million per month for GPU compute capacity, starting in October 2026 and running through June 2029 [1][2]. At full run rate, the deal is worth approximately $27.6 billion — nearly $30 billion when ramp-up payments are included [3]. The arrangement gives Google access to about 110,000 Nvidia GPUs and supporting hardware at SpaceX-operated data centers in Memphis, Tennessee [1][2].

The deal is remarkable not just for its scale but for its direction. Five years ago, SpaceX was the customer: a 2021 partnership placed Starlink ground stations inside Google Cloud data centers so SpaceX could sell satellite internet backed by Google's infrastructure [4]. Now Google is leasing compute from SpaceX. The reversal reflects both the desperate hunger for AI compute among hyperscalers and the unexpected emergence of SpaceX — a rocket company — as one of the world's largest operators of GPU data centers.

The Contract: Terms, Flexibility, and Risk Allocation

A Google Cloud spokesperson described the arrangement as "a short-term, timely agreement to ensure we have bridge capacity," noting that customer demand for its Gemini Enterprise AI services exceeded internal forecasts [1]. The contract includes several protective mechanisms for Google:

• Ramp-up pricing: A lower rate applies through September 2026 as SpaceX brings capacity online [1].
• Delivery deadline: SpaceX must have the full 110,000-GPU cluster operational by September 30, 2026. If it misses that date, Google gets a one-month grace period, after which it can terminate the agreement entirely or accept a reduced GPU count at a proportionally reduced monthly rate [2][3].
• Mutual termination: After December 31, 2026, either party can exit with 90 days' notice [1][2].

The 90-day termination clause makes the deal far more flexible than a typical multi-year infrastructure lease. Google is not locked in for three years — it can walk away after roughly 14 months with three months' warning. This protects Google from long-term dependency, but it also means SpaceX cannot count on the full $30 billion as guaranteed revenue.

No comparable cloud deal of this magnitude has been publicly disclosed. For context, Alphabet's total capital expenditure for 2026 is projected between $175 billion and $190 billion [5][6], so the ~$11 billion annual cost of the SpaceX lease represents roughly 6% of Alphabet's total capex budget.

Source: Company earnings reports

Data as of Jun 1, 2026CSV

How a Rocket Company Became an AI Landlord

The Memphis data centers at the center of this deal were not built for Google. They were built for xAI, Elon Musk's artificial intelligence company, which merged with SpaceX in February 2026 in a transaction that valued the combined entity at $1.25 trillion [7]. The flagship facility — Colossus — launched in September 2024 at a former Electrolux manufacturing site in South Memphis to train xAI's Grok language model [8].

Colossus has since expanded into what is currently the world's largest AI training cluster. The facility houses over 555,000 Nvidia GPUs across H100, H200, and next-generation GB200 accelerators, with total power capacity approaching 2 gigawatts [8]. The expansion cost approximately $18 billion [8].

After the xAI merger, SpaceX began leasing excess capacity to outside clients. In May 2026, Anthropic signed a deal to use the full capacity of Colossus 1 — more than 220,000 GPUs and over 300 megawatts of power — for $1.25 billion per month through May 2029 [9][10]. The Google deal, signed weeks later, covers a separate allocation of 110,000 GPUs.

Source: SpaceX S-1 Filing, SEC

Data as of Jun 5, 2026CSV

Combined, the Anthropic and Google contracts generate over $2.17 billion per month in compute revenue for SpaceX — more than $26 billion annually. Against SpaceX's 2025 total revenue of $18.7 billion [11], these two deals alone would more than double the company's top line if they run at full rate through 2027.

Source: SpaceX S-1 Filing

Data as of Jun 5, 2026CSV

The IPO Factor

The timing is not accidental. SpaceX filed its S-1 on May 20, 2026, targeting what would be the largest IPO in history: a Nasdaq debut at a valuation between $1.75 trillion and $2 trillion, with plans to raise approximately $75 billion [11][12]. The Google contract was disclosed in Amendment No. 2 to that filing, dated June 3 [2].

For potential IPO investors, the Google deal signals that SpaceX's AI infrastructure business has enterprise-grade customers willing to commit at scale. SpaceX's AI segment generated $3.2 billion in revenue in 2025 but lost $1.24 billion at the EBITDA level [11]. The Anthropic and Google deals could flip that segment to profitability — a critical narrative for a company seeking a $2 trillion valuation.

But the S-1 also discloses the AI segment's capital intensity: SpaceX spent $10.1 billion in total capex in Q1 2026 alone, with $7.7 billion directed toward AI and infrastructure [3]. Building data centers at this pace requires massive upfront investment, and the Google deal helps justify that spend by providing a contracted revenue stream.

Governance and Conflict-of-Interest Concerns

Elon Musk simultaneously serves as CEO, CTO, and chairman of SpaceX. Under the company's dual-class share structure disclosed in the S-1, Musk holds Class B shares carrying ten votes each, giving him approximately 79% voting control despite owning 42% of the company's equity [13]. The charter stipulates that only Class B shareholders — effectively Musk alone — can remove the CEO, chairman, or board members [13].

SpaceX has also adopted mandatory binding arbitration for all shareholder disputes, blocking investors from filing federal lawsuits including class actions [13]. A 3% ownership threshold for derivative suits means that at SpaceX's expected valuation, only Musk would meet the threshold to sue directors — the same directors he appoints [13].

These governance provisions are relevant to the Google deal because SpaceX's commercial arrangements sit within an ecosystem of Musk-controlled entities. The Memphis data centers were originally xAI facilities. xAI merged into SpaceX. Musk also runs Tesla, which has its own AI compute needs and Dojo supercomputer program. The question of whether compute capacity is allocated across Musk's companies on arm's-length commercial terms, or according to Musk's discretion, is one that SpaceX's governance structure does little to address.

No public reports indicate that Google's board or institutional shareholders have formally objected to the deal. However, the structural concerns are the kind that proxy advisory firms and institutional investors — particularly California pension funds, which have already criticized the dual-class structure — tend to flag during IPO roadshows [13].

As of this writing, neither the FTC, DOJ, nor EU competition authorities have announced any review of the Google-SpaceX computing agreement [14]. The deal is structured as a services lease rather than an acquisition or equity investment, which places it outside standard merger review thresholds. However, the sheer scale of bilateral dependency — Google paying SpaceX over $11 billion annually while SpaceX depends on Google for a significant share of its revenue — could attract scrutiny if regulators take a broader view of market concentration in AI infrastructure.

What Google Lacks — and Why It's Buying

Google is not short on data center expertise. Alphabet plans to spend up to $190 billion on capital expenditures in 2026, nearly doubling the $91 billion it spent in 2025 [5][6]. Its cloud backlog surged to $240 billion by the end of Q4 2025, more than doubling year-over-year [6]. So why lease GPUs from a rocket company?

The answer is time. Building a data center from scratch takes 18 to 24 months. Securing Nvidia GPU allocations requires placing orders years in advance. Google's own spokesperson framed the SpaceX deal as "bridge capacity" to cover a gap between surging Gemini Enterprise demand and the buildout of Google's own facilities [1]. SpaceX, having built Colossus at speed driven by Musk's urgency to train Grok, had capacity available now.

This dynamic — demand outrunning supply — is playing out across the industry. Global cloud providers are projected to invest a record $830 billion in capex in 2026, a 79% increase over 2025, driven almost entirely by AI infrastructure needs [5]. The GPU shortage has created a secondary market where companies with excess capacity can lease it at premium rates to those who need it immediately.

Competitive Implications for Satellite Broadband

While the current deal centers on GPU compute rather than satellite connectivity, it deepens the commercial relationship between Google and SpaceX in ways that affect the broader satellite broadband market. Google and SpaceX have maintained a partnership since 2021, when Starlink ground stations were deployed at Google Cloud data center properties [4]. Google was also an early investor in SpaceX.

For SpaceX's satellite broadband competitors, the concern is that these layered commercial relationships create an ecosystem that is difficult to replicate:

• Amazon's Project Kuiper (rebranded as Amazon Leo) plans to increase capex by $1 billion in 2026 with 20 launches scheduled, but its 198-satellite constellation is far smaller than Starlink's fleet of over 6,000 [15].
• Telesat Lightspeed is preparing to launch its first pathfinder LEO satellites this year, with commercial service planned for 2027 [15].
• Viasat continues to operate primarily through geostationary satellites, with higher latency than LEO constellations [15].

None of these competitors have an AI compute business to cross-subsidize their satellite operations. SpaceX's ability to generate billions in monthly data center revenue creates a financial moat that pure-play satellite operators cannot match.

The Case For and Against the Deal

The case for alarm: Two of the most capital-intensive private companies in the world are creating a bilateral dependency worth tens of billions of dollars, governed by a corporate structure that concentrates near-total control in a single individual who is also a political figure with influence over federal contracting and regulation. SpaceX's governance makes it nearly impossible for outside shareholders to challenge related-party transactions or allocation decisions across Musk's portfolio of companies. The 90-day termination clause, while protecting Google's optionality, also means SpaceX could find itself with a massive stranded-asset problem if Google walks away.

The case for the deal: Google needed GPU capacity that it could not build fast enough. SpaceX had it. The contract terms include meaningful protections — delivery deadlines, pro-rata pricing adjustments, and mutual termination rights. Starlink has demonstrated measurable enterprise-grade performance, with Gen3 satellites delivering 25.7-millisecond median latency in the United States, competitive with fixed broadband [16]. The Colossus facility is operational, not speculative. And the broader effect of the AI infrastructure buildout — including SpaceX's stated plans to develop orbital data centers [17] — could accelerate internet access in rural and underserved regions globally, a public-interest outcome that satellite broadband has long promised but underdelivered.

Oracle has already integrated Starlink into its enterprise communications platform across more than 100 countries [18]. Several organizations in the UAE region failed over to Starlink as backup connectivity after AWS data center issues in March 2026 [16]. These are real-world demonstrations of enterprise reliability, not theoretical projections.

What Comes Next

SpaceX's IPO roadshow is expected in the coming weeks, with a target listing date between June 18 and June 30 [12]. The Google deal — and the Anthropic deal before it — will be central to SpaceX's pitch that its AI infrastructure business is a durable revenue source, not a one-time windfall from excess capacity.

For Google, the key question is whether this remains bridge capacity or becomes structural dependency. If Alphabet's own data center buildout stays on schedule, the 90-day termination option gives Google a clean exit by early 2027. If demand continues to outstrip supply — as it has for the past two years — Google may find itself renewing or expanding the arrangement.

The broader industry signal is clear: the AI compute shortage has grown severe enough that a company spending $190 billion per year on infrastructure still needs to lease capacity from outside providers. Whether that shortage is a temporary bottleneck or a permanent feature of the AI era will determine whether deals like this become routine — or whether the Google-SpaceX agreement is remembered as the high-water mark of a supply crunch that eventually resolved itself.