xAI, SpaceX, and the Race for AI Buildout

Last updated: August 7, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely

xAI, SpaceX, and the race for AI buildout became one story instead of three on February 2, 2026. Elon Musk merged the two companies into a single, roughly $1.25 trillion entity, and the combined company then raised its own $20 billion war chest for xAI's models, expanded a 555,000-GPU data center in Memphis, and filed with the FCC to put up to a million AI-compute satellites in orbit.

Short answer: xAI merged into SpaceX on February 2, 2026, forming a roughly $1.25 trillion company. xAI still closed its own $20 billion Series E at a $230 billion valuation in January 2026, funding Colossus 2's 555,000 GPUs in Memphis. SpaceX then filed with the FCC for up to 1 million orbital data-center satellites, with the first two prototypes targeted for early 2027.

ChatGPT homepage — screenshot of chatgpt.com
ChatGPT homepage — screenshot of chatgpt.com

I review AI tools for a living, and Grok is one of the products I test regularly. This deal isn't background noise to me — it's a corporate restructuring of the company behind a chatbot in my testing rotation. When I went back to the primary filings instead of the "Musk merges AI with rockets" headlines, the story split into three separate buildouts happening at once. Each has its own money, its own timeline, and its own risk.

What you'll need to follow this story

You don't need a finance background, just three habits. Keep the merger (corporate structure) separate from the Series E (xAI's own funding round) and separate again from the FCC filing (a regulatory ask, not an approved product). Treating those as one event is where most coverage goes wrong. Bookmark the FCC's ECFS filing search and the House Energy and Commerce Committee's press page. Both publish primary documents faster than most news aggregators pick them up. And if you use Grok yourself, keep xAI’s own pricing docs open. None of this buildout news has moved what you pay for the product yet.

Step-by-step: how to make sense of the buildout

1. Start with the merger, not the money

On February 2, 2026, SpaceX announced it had acquired xAI, folding Musk's AI lab into the rocket company at a combined valuation of roughly $1.25 trillion, according to Via Satellite’s report the same day. That's a corporate-structure event: it changes who owns what, not how much compute exists yet.

2. Check what xAI actually raised on its own

Weeks earlier, on January 6, 2026, xAI closed a separate $20 billion Series E at a $230 billion valuation. Backers included Nvidia, Cisco, Fidelity, Qatar Investment Authority, MGX, and Tesla, which put in roughly $2 billion subject to approvals, per CNBC’s reporting. That money is earmarked for Grok training and the Memphis data centers, not the merger itself.

3. Separate ground compute from orbital compute

xAI's Colossus 2 site in Memphis runs on hardware that exists today: 555,000-plus Nvidia GPUs across roughly 2 gigawatts of combined capacity. The GPU cost alone is near $18 billion, and the total build cost runs $35–40 billion once buildings, cooling, and backup power are included, per Introl’s January 2026 breakdown. The orbital plan, internally called Starmind, is a regulatory filing for up to 1 million satellites. Nothing is in orbit yet.

4. Read the permitting fight as its own thread

Colossus runs partly on gas turbines that Memphis-area regulators say were installed without full air permits. On July 29, 2026, Rep. Frank Pallone, ranking member of the House Energy and Commerce Committee, sent SpaceXAI a letter demanding a tour and emissions records. The letter cites an analysis that the turbines are the single largest pollution source in the Memphis area — five times as polluting as the next-largest source, Memphis International Airport — per the committee’s own press release. This is a compliance story, running parallel to the buildout, not a part of it.

5. Note what's confirmed versus what's a target

A prototype AI1 satellite, roughly 70 meters tip-to-tip and delivering about 150 kilowatts of peak compute power, is the near-term reality: two are slated to launch in early 2027. SpaceX's own stated goal — roughly 1 gigawatt of orbital compute per year by late 2027, with commercial service starting around 2028 — is a projection, not a shipped product.

Example prompts you can copy

Use these with any AI chat tool to pull the current state of this story instead of trusting a headline:

Summarize the most recent SEC or FCC filings from SpaceX or xAI related to data centers, satellites, or corporate structure. List the filing date, filing type, and a one-sentence summary of what changed.

xAI closed a $20 billion Series E in January 2026 at a $230 billion valuation. Has xAI raised additional funding since then, and if so, from whom and at what valuation? Cite your source and its publish date.

Has the House Energy and Commerce Committee or any state regulator issued a follow-up statement on SpaceXAI's Memphis, Tennessee data center turbine permits since July 29, 2026? Summarize what's changed and cite the source.

These work because they point the model at specific filings and dates instead of asking it to explain "the AI buildout race" in general, which is where answers get vague and stale fast.

Common mistakes to avoid

In my testing of how this story gets reported and re-reported, the most common mistake is treating the $1.25 trillion merger valuation and xAI's $230 billion Series E valuation as the same number. They're not. One is the combined SpaceX-plus-xAI entity; the other was xAI alone, three weeks earlier, before the merger closed. The second mistake is assuming the 555,000 GPUs at Colossus 2 represent xAI's full build-out. That's roughly 55% of the company's stated 1-million-GPU target, and the rest isn't built yet. Third, people read "SpaceX filed with the FCC for 1 million satellites" as if the satellites exist. A filing is a regulatory request, and the two AI1 prototypes aren't scheduled to launch until early 2027. Fourth, don't assume the buildout has touched Grok's price yet. When I checked xAI's own pricing docs while researching this piece, SuperGrok was still $30 a month (or $300 a year), unchanged from what I confirmed for my Grok how-to guide a few days earlier.

Ground vs. orbital: how the two buildouts compare

Buildout layer Compute scale Cost Status as of August 2026 Biggest open risk
Colossus 1 & 2 (Memphis, ground) ~2 GW combined; 555,000+ Nvidia GPUs at Colossus 2 alone ~$18B in GPUs; ~$35–40B total build-out Operating; a third building ("Macrohardrr") targeted for Q2–Q3 2026 69 gas turbines reported still unpermitted; Congress demanding records and a site tour
Terafab (Austin, chip fab) Starting near 100,000 wafer starts/month, scaling toward 1,000,000/month $20–25B, funded jointly by Tesla, SpaceX, and xAI Announced; no public production start date Unproven timeline for reaching full wafer-start capacity
Starmind / AI1 (orbital) Two AI1 prototypes at ~150 kW peak compute each; FCC filing covers up to 1,000,000 satellites Not disclosed separately from SpaceX's broader capex Prototypes targeted for early 2027; commercial service targeted 2028 Musk's 2–3 year cost-parity claim vs. independent analysts projecting "well into the 2030s"

The pattern in that table is the story in miniature: the ground buildout is real, expensive, and already drawing regulatory scrutiny, while the orbital buildout is a well-funded bet that won't have a single satellite in the sky before 2027.

Tools that make this easier

If you're trying to track what any of this changes for the products you use day to day, my how-to guide to Grok and Grok vs. ChatGPT comparison cover what's shipped and what a subscription buys right now. My Grok 4.5 pricing breakdown goes deeper on the API side if you're building on top of it. The financing side of this same story is covered in my piece on why the AI trade increasingly runs on borrowed money and in Morgan Stanley’s debate over how much of SpaceX’s valuation is AI, a piece written before this merger that's a useful before-and-after read. The power and permitting angle behind Colossus's turbine fight is the same dynamic I mapped state by state in AI data centers are driving up power bills. Nvidia’s $750 billion in AI deals covers the same financing pattern showing up across the rest of the industry, not just at xAI. My AI tool ratings hub is where I keep Grok's score updated as the product changes, separate from any of this corporate news.

My take

The merger is the least interesting number in this story. $1.25 trillion is a combined-entity valuation, and the actual work is happening in three separate places at three separate speeds. Colossus is real and running today, which is also why it's the one drawing a congressional letter over unpermitted turbines. Terafab is a serious commitment on paper with no public production date yet. Starmind gets the most attention and has the least evidence: a filing and a prototype timeline, built on a cost-parity claim that Musk puts at two to three years and outside analysts put a decade further out. I'd watch the Memphis permitting fight first, since it has real regulatory deadlines. I'd treat the orbital number as a target until an AI1 prototype is actually in the sky.

Frequently Asked Questions

Did SpaceX buy xAI, or did they merge as equals?

SpaceX acquired xAI, folding it into the rocket company as a single entity, according to the February 2, 2026 announcement. It's structured as an acquisition, not a merger of equals. Musk and outlets covering the deal have still used "merger" loosely to describe the combined roughly $1.25 trillion company.

Is the $1.25 trillion valuation the same as xAI's $230 billion valuation?

No. The $230 billion figure came from xAI's own Series E, closed January 6, 2026, three weeks before the SpaceX acquisition. The $1.25 trillion figure values the combined SpaceX-plus-xAI entity after the deal, so the two numbers aren't measuring the same thing.

Are SpaceX's orbital data centers actually built yet?

No. As of August 2026, SpaceX has an FCC filing requesting authority for up to 1 million satellites and two AI1 prototype satellites scheduled to launch in early 2027. Commercial orbital service isn't targeted until around 2028.

Does any of this affect what I pay for Grok?

Not so far. When I checked xAI's pricing docs while researching this piece, SuperGrok was still $30 a month (or $300 a year) and the free tier was unchanged — the buildout news sits at the infrastructure and corporate-structure layer, several steps removed from the subscription price.

Why is Congress involved in a data center story?

Because of power, not AI policy. The House Energy and Commerce Committee's July 29, 2026 letter targets gas turbines powering Colossus that regulators say were installed without full air permits, which is an environmental-compliance issue that happens to sit underneath an AI data center rather than an AI-specific dispute.