One day’s posts, two very different time scales. In ENN’s previous article in this thread, Elon Musk set out SpaceX’s long-range aim of 1TW a year of capacity from Earth, rising to over 1PW a year from the Moon1. What is new now is a near-term number. In a separate reply on 11 November, Musk wrote that SpaceX has a chance of reaching about half a gigawatt of “actually usable” compute next year2.
He also explained why he chose that wording, and he put a number on how fast SpaceX turns raw hardware into a working system: about two months of round-the-clock effort for a large cluster2.
@zerohedge SpaceX has a chance of going about half the GW next year of actually usable compute. “Actually usable” is an important metric. It takes us ~2 months working 24/7 to get a large compute cluster working reliably. I’d guess closer to 6 months for others.
— @elonmusk Nov 11, 2026
What Musk said
Replying to @zerohedge, Musk wrote:
SpaceX has a chance of going about half the GW next year of actually usable compute. “Actually usable” is an important metric. It takes us ~2 months working 24/7 to get a large compute cluster working reliably. I’d guess closer to 6 months for others.
That is the full post2. It contains three claims worth separating. First, the target: about half a gigawatt of compute next year, framed as “a chance” rather than a promise2. Second, the yardstick: compute that is “actually usable,” which he calls “an important metric”2. Third, the speed: roughly two months of 24/7 work for SpaceX to get a large cluster working reliably, against his guess of closer to six months for others2.
The careful phrasing matters. Musk does not say SpaceX will hit half a gigawatt. He says it has a chance. ENN reports it in exactly those terms.
Why “actually usable” is the number he picked
Big AI clusters are often described by the power they can draw or the chips they hold. Musk’s post points at a different measure: how much of that capacity is working reliably2. His comment about time to reliability shows why he treats the two as separate. By his account, even SpaceX needs about two months of continuous work after a large cluster is assembled before it runs reliably, and he estimates others take about three times as long2.
If his estimate holds, that speed gap is a competitive edge in its own right. Two months versus six months is the difference between hardware earning its keep in one quarter and hardware sitting partly idle for half a year. The post itself does not say how SpaceX achieves that pace or which clusters the figure covers, so ENN does not speculate on the method2.

The case for compute in space
The same day, Musk restated the argument that ties this thread together: why put compute in space at all. Replying to @TheFlintFace and @IterIntellectus, he wrote that “100GW in space is roughly equal to 800GW on Earth”3.
His reasoning was about power, not chips. He wrote that “solar on Earth must be paired with massive batteries to continue powering compute at night and supplemented with combustion or nuclear power during winter,” and concluded: “These are not the same at all”3.
Put simply, a ground-based solar farm has to be built far larger than its headline rating, and backed by storage and other generation, to keep a data center running every hour of every season. Musk’s claim is that capacity in space does not carry that burden, which is how he arrives at his roughly eight-to-one comparison3. The ratio is his own estimate; the post does not show the calculation behind it.
That framing explains how the near-term and long-term numbers relate. Half a gigawatt of usable compute next year is the step in front of SpaceX2. The 1TW-a-year Earth goal and the 1PW-a-year Moon goal are the horizon1. In his 1TW post, Musk added that if SpaceX achieves that outcome, “SpaceX will obviously be worth more than Earth, because space is so staggeringly enormous!”1

Where SpaceX’s compute business stands today
SpaceX is not starting from zero. According to AOL, a February 2026 all-stock merger brought xAI, along with the X social platform, under the SpaceX umbrella, turning the company into one of the world’s largest AI infrastructure landlords4.
AOL reports that Anthropic agreed to pay $1.25 billion per month through May 2029 for full access to the Colossus 1 data center near Memphis, Tennessee, a facility housing more than 220,000 NVIDIA GPUs and drawing 300 megawatts of power4. Google signed a similar agreement on June 5 to pay $920 million per month from October 2026 through June 2029 for access to roughly 110,000 NVIDIA GPUs4. Combined, AOL says, the two contracts add about $26 billion in annual run-rate revenue4.
Those figures give a sense of scale. A single facility, Colossus 1, draws 300 megawatts, according to AOL4. Musk’s half-gigawatt outlook is a measure of usable compute across SpaceX next year2; his post does not say which sites it would come from or whether it counts capacity on the ground, in orbit or both.
The business has grown fast on the public markets too. AOL reports that SpaceX listed on Nasdaq as SPCX on June 12, priced at $135 per share and raising $75 billion in the largest IPO in history4. Its first post-IPO report showed $7.81 billion in Q2 2026 revenue, up 92% year over year4.
Costs and open questions
Independent coverage also lays out the price of building at this pace. AOL reports that SpaceX’s first post-IPO filing disclosed $28.5 billion in first-half capital spending and about $25 billion in negative free cash flow, which rattled investors4. The AI segment, inherited through the xAI merger, lost $6.355 billion in 2025 on $3.2 billion in revenue, according to the same report4.
AOL also notes that both compute rental agreements include 90-day termination clauses that either party can exercise after December 31, 20264. In a separate reply on 11 November, to @aaronburnett, Musk wrote: “We would be happy to take the compute back. No problem. But I think the opposite is going to happen.”5 The post does not spell out which compute or which arrangement he was referring to, so ENN reports his words without attaching them to a specific deal.
The broad picture is a company spending heavily now on capacity it expects to keep putting to work, led by a chief executive who says the hard part is not buying hardware but making it reliably useful, quickly24.
What to watch next
- The half-gigawatt mark. Whether SpaceX reports usable compute near the level Musk described for next year2.
- Time to reliability. Any further detail on how SpaceX brings large clusters online in about two months2.
- Space-based capacity. Concrete steps toward the 1TW-a-year Earth goal and the 1PW-a-year Moon goal1, and toward the power advantage Musk claims for compute in space3.
- The rental contracts. How the Anthropic and Google agreements develop after the 90-day termination window opens at the end of 20264.
For now, the evidence is Musk’s own word on X. It shows his outlook and his reasoning, not built or audited capacity. ENN will update this thread as SpaceX’s compute numbers come into view.



