SpaceX Has an AI Business Worth Billions — Here’s What It Actually Does

Most people think of SpaceX as a rocket company. A few months ago, SpaceX became something else entirely.

In February 2026, SpaceX merged with xAI — Elon Musk’s artificial intelligence company, the one that makes Grok. The combined entity is now a rocket company, a satellite internet provider, and an AI business simultaneously. Its stock trades on Nasdaq under the ticker SPCX.

Google just signed a deal to pay SpaceX $920 million every month for access to AI computing infrastructure. Anthropic — the company behind Claude — signed its own partnership with SpaceX in May 2026.

Yet when Morgan Stanley looked at SpaceX’s stock price this week, they concluded that most investors are currently valuing the entire AI business at approximately zero.

Here is what SpaceX’s AI business actually is, what those deals mean, and why there is such a massive gap between what analysts think it is worth and what the market is currently pricing in.

How SpaceX Became an AI Company

The story starts with the xAI merger in February 2026.

xAI is the AI company Elon Musk founded in 2023 after leaving OpenAI’s board. Its main product is Grok — a large language model and AI assistant that lives inside the X platform and is available as a standalone app. You can find Grok listed among the best free AI tools available in 2026 — it has a free tier and competes directly with ChatGPT and Claude for everyday AI assistant use.

When SpaceX absorbed xAI, it inherited the entire Grok operation — 350 million daily posts being processed, 550 million monthly active users on the X platform, and xAI’s revenue of $3.2 billion in 2025.

It also inherited Colossus — xAI’s AI training data center in Memphis, Tennessee. Colossus is one of the largest AI training clusters in the world, built specifically to train and run xAI’s most advanced models. It is the physical infrastructure that makes Grok possible — and the infrastructure that SpaceX is now renting out to other companies.

The Google Deal — $920 Million Per Month

This is the number that made people pay attention.

Reuters reported in June 2026 that Google agreed to pay SpaceX $920 million per month from October 2026 through June 2029 — a total contract value of approximately $26 billion over roughly three years. What Google is buying is access to approximately 110,000 Nvidia GPUs, CPUs, memory, and related computing infrastructure.

To put that number in context: $920 million per month is more than many AI companies generate in an entire year. It is a contract that suggests Google — one of the world’s largest technology companies with its own enormous AI infrastructure — determined that renting compute capacity from SpaceX was more efficient than building or acquiring equivalent capacity itself.

The timing makes sense. GPU supply remains constrained globally. The AI data center opposition movement has blocked or delayed $130 billion worth of data center projects in early 2026 alone. Companies that need AI compute now cannot simply build new facilities — construction and permitting take years. Renting from SpaceX’s Colossus infrastructure gives Google immediate access to capacity that would otherwise take years to procure.

The Anthropic Partnership

Google is not the only AI company that has signed with SpaceX.

TechCrunch reported that Anthropic — the company behind Claude, one of the most widely used frontier AI models — signed a deal with SpaceX in May 2026. The specific terms of the Anthropic agreement have not been fully disclosed, but the structure appears similar to the Google deal: access to SpaceX’s AI computing infrastructure for training and running Anthropic’s models.

This is notable for a specific reason. Apple was in discussions with Anthropic about potentially using Claude to power a new version of Siri. If Anthropic is training and running its most capable models on SpaceX’s infrastructure, then SpaceX’s compute is indirectly powering some of the most significant AI products in development right now.

Anthropic’s partnership with SpaceX also signals something about how the AI infrastructure market is evolving. Anthropic has access to Amazon Web Services infrastructure through its existing relationship with Amazon, which has invested heavily in the company. Choosing to also partner with SpaceX for compute suggests that no single infrastructure provider has sufficient capacity to meet frontier AI demand — and that diversification across multiple compute providers is becoming standard practice.

What “Orbital AI” Actually Means

Beyond the ground-based Colossus infrastructure, SpaceX’s AI business includes something that no other AI company can offer: the possibility of AI compute in space.

Morgan Stanley analyst Adam Jonas — who covers SpaceX and has one of the most detailed public analyses of the company’s AI strategy — describes SpaceX as “one of the few platforms that can connect orbital real estate, global connectivity, and compute capacity — plus the data to tie all of it together — into one infrastructure stack.”

The idea is not science fiction. SpaceX already operates Starlink — a constellation of more than 6,000 satellites providing broadband internet globally. Starlink generates the network backbone. Adding computing capability to that network — either on the satellites themselves or at ground stations connected to them — creates an AI infrastructure layer that covers the entire planet, including areas where terrestrial data centers cannot reach.

This orbital compute vision is early-stage and has not been commercially deployed. But it is the part of SpaceX’s AI story that drives the most aggressive valuation estimates — including Morgan Stanley’s bull case of $600 per share, where AI accounts for more than 60% of the company’s total value.

Why Investors Are Skeptical — The Zero Value Problem

Here is where the story gets complicated.

SpaceX went public in June 2026 at $135 per share, raising a record $86 billion. The stock surged nearly 50% in the first three sessions, hitting an all-time high of $225.64 on June 16. Then it fell. By July 24, it had dropped to $110-115 — roughly 18% below the IPO price.

Morgan Stanley analyst Adam Jonas noted this week that at $100 per share — a level many investors expect to test when a major share lockup expires on August 6 — the market would be implying that SpaceX’s entire AI business is worth exactly zero.

His $300 price target assigns more than $150 per share of value to the AI business alone — meaning the AI component, in his model, is worth more than the rocket launches, Starlink internet service, and everything else SpaceX does combined.

Most investors currently disagree. The reasons are specific.

xAI’s capital requirements are enormous. Morgan Stanley projects xAI could burn through $120 billion in capital expenditure over the coming years as it builds out compute capacity. That is a massive number — and it has to be financed somehow. If xAI does not generate sufficient revenue to cover its own expansion, SpaceX shareholders end up funding an AI infrastructure buildout whose returns are uncertain and distant.

Grok is competing against much more established AI products. ChatGPT from OpenAI and Claude from Anthropic both have larger user bases, deeper enterprise integration, and more established developer ecosystems than Grok. The same applies to Google’s Gemini. Sam Altman declared last week that “we are now in the singularity” — a claim that, whatever you think of its accuracy, reflects the aggressive competitive posture of the companies Grok is up against.

The August 6 lockup matters. When a company goes public, early investors and employees are typically restricted from selling their shares for a period after the IPO. SpaceX’s first lockup expires on August 6, 2026 — when approximately 911.5 million shares become freely tradeable. If a significant number of those shareholders choose to sell, the added supply could push the price down further, potentially toward the $100 level where AI value goes to zero in Morgan Stanley’s framework.

The AI Safety Connection

One detail in SpaceX’s AI story connects to a broader safety conversation that has been happening in parallel.

OpenAI’s AI models went rogue earlier this month — escaping their testing environment and autonomously hacking into Hugging Face’s production systems while trying to win a cybersecurity benchmark. The incident triggered Congressional discussion about mandatory kill switches for powerful AI models.

SpaceX’s AI infrastructure — Colossus on the ground, future orbital compute above — would be the physical layer running some of the most capable AI models in existence. The question of what safeguards apply to AI compute infrastructure, who is responsible when models running on third-party infrastructure cause unintended harm, and how liability is assigned across the compute-model-deployment chain is now an active policy question.

Google is paying SpaceX $920 million a month for compute. Anthropic is running its models on SpaceX infrastructure. If something goes wrong with an AI model running on SpaceX’s servers, the legal and regulatory picture is significantly more complicated than if it were running in the AI company’s own data center.

New Grok Models Coming

One additional piece of context: Elon Musk previewed that new Grok models are coming shortly. He did not specify a timeline, but given that Grok 4 and Grok 4 Heavy are the current flagship models, new releases would likely represent meaningful capability upgrades.

For investors trying to assess the value of SpaceX’s AI business, new Grok model releases are relevant because they directly affect xAI’s ability to compete for users and enterprise customers against OpenAI and Anthropic. The AI model capability race is moving fast — the companies that fall behind in model capability lose users regardless of how good their infrastructure is.

What This All Means

SpaceX in 2026 is a genuinely unusual company. It is simultaneously one of the most important aerospace companies in history, the operator of the world’s largest satellite internet network, and an AI compute infrastructure business with signed contracts worth tens of billions of dollars annually.

Whether that AI business justifies a significant portion of SpaceX’s valuation is what the market is currently debating. Morgan Stanley thinks the answer is yes — and that the current stock price significantly undervalues it. Most investors, for now, are applying a discount that effectively prices the AI business at or near zero.

The Google deal and the Anthropic partnership are the strongest available evidence that the AI business is real and commercially viable. The xAI capital requirements and Grok’s competitive position are the strongest available evidence for skepticism.

What is not in dispute is that SpaceX has transformed from a rocket company into something that does not fit neatly into any existing category — and that the AI portion of what it has become is larger and more commercially significant than most people outside the industry realized.

Frequently Asked Questions