THE POST
Perspicacious path to a petawatt/year
ENGAGEMENT
- likes5,519
- reposts527
- replies678
- quotes100
- bookmarks323
- views2,171,550
AS RETURNED BY THE X API · FETCHED 2026-10-08 12:21 UTC
WHAT IT SAYS
A one-line post with no link, image or quoted post: Musk says there is a shrewd ("perspicacious") path to adding a petawatt, 10¹⁵ watts, per year. The post itself does not say what the path is.
CONTEXT
"Petawatt per year" matches language Musk and SpaceX have used all year for solar-powered AI compute in space. In a 2 February 2026 update on SpaceX acquiring xAI, Musk wrote that an electromagnetic mass driver and lunar manufacturing could "put 500 to 1000 TW/year of AI satellites into deep space, meaningfully ascend the Kardashev scale and harness a non-trivial percentage of the sun's power." 1,000 TW is one petawatt.
At a presentation at Giga Texas in March 2026, Musk unveiled Project TERAFAB, which the outlet describes as a plan to move power-hungry AI infrastructure to the lunar surface, with mass drivers launching from the Moon. The stated goal is to break the "terawatt ceiling" of Earth-bound computing and reach a "petawatt-scale era". The outlet notes that a lunar mass driver remains purely theoretical.
According to reporting on SpaceX's pre-IPO filing, the company describes lunar factories for AI compute satellites, a lunar mass driver, and a "petawatt-scale constellation of orbital AI compute" that it wants to start deploying as early as 2028. A former SpaceX engineer quoted in the same piece says Musk is "usually off by a few years and a little on the optimistic side".
The hardware that exists on paper today is SpaceX's AI1 "Starmind" satellite: up to 250 kW peak and 175 kW average compute payload, a 75 m wingspan and 75 kW per ton, with the Gigasat factory in Bastrop planned to produce thousands of them starting as soon as late 2027. SpaceX says it plans up to 1 million Starmind satellites.
Scale check (Observatory's arithmetic from the figures above): 1 PW is about 49 times humanity's current average power of roughly 20.6 TW, and about 1,400 times the 692 GW of renewable capacity the world added in 2025. At AI1's 250 kW peak and 75 kW per ton, one petawatt per year would mean about 4 billion such satellites and roughly 13 billion tonnes of spacecraft a year. Musk himself has put Starship's eventual delivery at 1 to 10 million tons per year to orbit, three to four orders of magnitude short. That gap is why the petawatt plan runs through lunar manufacturing rather than launches from Earth. A full 1-million-satellite Starmind constellation would be about 250 GW peak, roughly 1% of humanity's current power.
WHAT IT MEANS FOR K
Speculative. If a petawatt per year of solar-powered compute were ever deployed, K would jump: one year at that rate takes humanity from about 20.6 TW to about 1 PW, or K from 0.7313 to about 0.90 (Observatory arithmetic). Today it changes nothing measured: no Starmind satellite is in orbit, the first are planned from late 2027, and the lunar factory and mass driver exist only as plans.
CLAIMS
- A path to a petawatt of new capacity per year exists (the post).
- Lunar manufacturing plus a mass driver could deliver 500–1,000 TW/year of AI satellites (Musk, Feb 2026).
- SpaceX wants to start deploying a petawatt-scale orbital compute constellation as early as 2028 (pre-IPO filing, as reported).
EVIDENCE
- AI1 specifications are published: 250 kW peak, 75 kW/ton (SpaceX).
- A factory for thousands of AI satellites is planned from late 2027 (SpaceX).
- An FCC filing for up to 1 million Starmind satellites has been accepted, which is about 250 GW peak if fully built (Teslarati; Observatory arithmetic).
- No lunar mass driver or lunar factory has been built or tested at scale (Interesting Engineering).
REASONING
- Power harnessed in orbit counts toward K in Kardashev's sense, but this site's K tracks Earth's primary energy, so orbital solar would need to be added as a separate line if it ever becomes material.
- Even the near-term, specified part of the plan (1 million AI1 satellites, about 250 GW) would add about 1% to humanity's power, which is real but small on a log scale (about +0.0005 K).
- The petawatt figure depends on infrastructure that has not been demonstrated, so it is treated as a claim, not a forecast.
SEE ON THIS SITE
TAGS
- ENERGY
- SPACE
- COMPUTE
- SOLAR