SpaceX doubles down on orbital compute as markets wobble and the AI capex narrative strains

SpaceX has told investors and partners it intends to put orbital AI computing infrastructure into space by late 2027, with initial demonstrations of in-orbit inference and training workloads targeted for the back half of that year, according to a Reuters dispatch relayed by Cointelegraph on 10 June 2026 at 00:07 UTC. The disclosure lands the same week US equities shed close to $2 trillion in market value over a roughly two-hour window on 9 June 2026, and one day after Anthropic moved to release Claude Fable, a safeguarded frontier-scale model that the company and several prediction markets are calling a "Mythos-class" system.

The convergence of those three events is the story. Each one is significant in isolation. Read together, they describe a market that is simultaneously repricing the AI capex supercycle, betting that the next leg of compute will move off-planet, and watching the public markets twitch on every new model release. Whether the late-2027 timeline is realistic is a separate question from whether the direction of travel is real. On the latter, the answer is increasingly obviously yes.

The SpaceX disclosure

The Reuters report, summarised by Cointelegraph at 00:07 UTC on 10 June, frames SpaceX's orbital compute ambition as a complement to, rather than a replacement for, terrestrial hyperscale data centres. The plan is to fly dedicated compute payloads — inference accelerators and supporting thermal and power systems — on Starship, taking advantage of the launch cost curve that SpaceX itself has bent. Initial demonstrations are targeted for late 2027. The bet is straightforward: continuous solar irradiance in orbit, the option of radiative cooling into deep space, and a launch cadence that is already a competitive moat.

Two things are worth flagging. First, the timeline is aggressive. Late 2027 is roughly eighteen months out, and the regulatory and spectrum-coordination work for transmitting AI workloads to and from orbit at production scale has not been publicly detailed. Second, the disclosure arrives while SpaceX itself is preparing for a public listing that Reuters reporting, cited by Unusual Whales on 9 June 2026 at 21:58 UTC, says is already roughly four times oversubscribed. Whether the orbital-AI roadmap is a core pillar of the IPO narrative or a separate optionality story is a question the S-1, when it lands, will answer more clearly than the company has so far.

A $2 trillion hour

The market context matters. On 9 June 2026 at 17:00 UTC, Cointelegraph flashed an alert that US stocks had lost close to $2 trillion in market value over a roughly two-hour window. Cointelegraph did not attribute a single trigger in that alert, and the underlying tape is consistent with a leverage flush rather than a fundamental repricing of any one name. But the timing is hard to ignore: a leveraged risk-off move, a flagship IPO already multiple times oversubscribed, and a frontier model release from one of the most-watched AI labs in the same trading week.

The reasonable read is that the market is no longer treating AI capex as a one-way bet. Even with a $2 trillion drawdown, the structural argument has not collapsed. It has, however, started to require justification per dollar, per gigawatt, and per shipped token. That is the regime change.

Claude Fable and the Mythos tier

Anthropic, meanwhile, is shipping a new flagship. Per The Information, cited by Cointelegraph at 14:33 UTC on 9 June 2026, the company released Claude Fable — branded as a "Mythos-class" model, with safeguards described as the company's most extensive to date. Polymarket independently confirmed the launch framing the same day at 14:26 UTC, with a market positioning the model in the frontier tier.

The naming is the story. "Mythos-class" is Anthropic's own tier, not a standard benchmark category, and the company has used the term to signal a step up in capability and a step up in mitigation work. That matters because the capex supercycle is increasingly tied to the assumption that each new model generation will pull forward demand for inference. If that assumption holds, SpaceX's orbital compute bet has a customer. If it doesn't, late 2027 looks further away than eighteen months.

Stakes and the road to late 2027

The structural frame is plain: compute is moving up the stack, and the next constraint is power, cooling, and real estate. The companies that solve those constraints on Earth get the next two years; the companies that solve them off it get the decade after that. SpaceX is positioning for the second race while still being judged on the first.

What remains uncertain is the actual technical envelope. The Reuters report, as relayed by Cointelegraph, does not specify power output per satellite, inter-satellite linking topology, or the customer pipeline beyond internal SpaceX and Starlink-adjacent workloads. The $2 trillion drawdown is reported in headline terms; the underlying concentration of losses across mega-cap tech versus the broader index is not detailed in the source material. And "Mythos-class" is, at this point, a vendor label. The benchmarks that determine whether Claude Fable sits in the same tier as the previous generation's frontier models will come from third parties in the weeks ahead.

What is not in doubt is the direction. Compute is leaving the basement. Some of it is going into orbit. The market is starting to price that, and the wobbles along the way are the cost of getting there.

— Monexus framed this as a single convergence story — orbital compute, frontier models, and a leverage flush in the same week — rather than three disconnected wires, because the capex assumptions tying them together are what the next eighteen months will actually test.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/s/cointelegraph
• https://t.me/s/cointelegraph
• https://t.me/s/cointelegraph
• https://t.me/s/cointelegraph