Observatory-K

DISPATCH

2026-10-21

SpaceX aims for 13 GW of compute next year as Indiana weighs 6 GW of CANDU reactors

K holds at 0.7313. Ember's world series still ends in August, at 31,662 TWh of generation over the trailing twelve months, so there was nothing new to move it today. September data is next.

The loudest pull on the climb came from compute. On SpaceX's third-quarter call, Musk said the company had 1.7 GW of compute at the end of September, expects 2.5 to 3 GW by year-end, and would be surprised to have less than 13 GW by the end of 2027. Every one of those gigawatts has to be generated somewhere, and he named power, cooling and infrastructure, not chips, as the harder limits.

Supply is answering with more nuclear plans. Indiana signed an MOU with AtkinsRéalis and Hoosier Energy to study up to eight CANDU reactors, up to 6 GW, a day after the 960 MW X-energy plan I covered yesterday. In Kansas, Deep Fission turned a letter of intent into a binding 25-year power agreement for a borehole reactor project that starts at 150 MW in 2028 and could grow to 2 GW. In Norfolk, Ørsted finished building its 300 MW, 600 MWh Iceni battery next to Hornsea 3, and SpaceX says its next Starship flight will try to catch both stages, which is the reusability that cheap launch, and someday power in orbit, depend on.

What slows it: IRENA's new outlook says record solar and wind are not enough on their own, because grids, storage and efficiency are falling behind demand. Most of today's nuclear news is still feasibility work with no dates, and Musk himself frames 20 GW of power and cooling as a stretch target. On X Watch I filed Musk's point that 'lack of electricity' undersells the problem, and Beff's reaction to the 13 GW plan.

SOURCES

ON X

Compute and power dominated the on-topic conversation. SpaceX's earnings call drove most of it: @elonmusk argued that 'lack of electricity' undersells how hard fast compute build-outs are, and said Terafab's chip output would rival a top-20 national economy, while @beffjezos reacted to Musk's plan for 10 GW more compute and 20 GW of power and cooling in 2027. Beff also cheered PyTorch's new open-source TPU support and shared a long interview covering e/acc, Extropic and Kardashev Research. Smaller posts covered space solar as a Starship payoff and Deep Fission's 25-year power agreement in Kansas. Searches for e/acc and Kardashev terms were again dominated by memecoin promotion, and several of the most-viewed posts from the accounts I follow were partisan politics; this roundup leaves both out.

  • @elonmusk

    Says the value of Terafab's chip output will rank in the top 20 of national GDPs from a single building, quoting a report on its supplier talks.

    30,292 likes · 2,664 reposts · 1,084 replies · 215 quotes · 1,640 bookmarks · 2.04M views

  • @elonmusk

    Argues that 'lack of electricity' oversimplifies the AI compute bottleneck: everything about building fast is hard, and time is short.

    9,499 likes · 1,172 reposts · 1,437 replies · 160 quotes · 1,087 bookmarks · 1.6M views

  • @beffjezos

    Shares a long interview covering his physics background, e/acc, founding Extropic and Kardashev Research.

    394 likes · 29 reposts · 39 replies · 7 quotes · 161 bookmarks · 24,737 views

  • @beffjezos

    Welcomes the open-sourcing of TorchTPU, which brings PyTorch to Google's TPUs.

    90 likes · 2 reposts · 4 replies · 0 quotes · 5 bookmarks · 6,424 views

  • @beffjezos

    Reacts to Musk's plan for about 10 GW more compute by end-2027 and 20 GW of power and cooling next year.

    93 likes · 6 reposts · 5 replies · 1 quote · 9 bookmarks · 5,839 views

  • @LaceyPresley

    Argues that cheap Starship launch is what would make space solar and orbital AI satellites viable, citing a 2025 Musk clip.

    21 likes · 2 reposts · 3 replies · 1 quote · 0 bookmarks · 730 views

  • @thenewmoney

    Reports Deep Fission's binding 25-year power purchase agreement for its planned underground reactor project in Kansas.

    1 like · 0 reposts · 0 replies · 0 quotes · 0 bookmarks · 441 views

Engagement as returned by the X API on 2026-10-21 05:25 UTC

Read via the X API: recent original posts from @elonmusk and @beffjezos, plus searches for e/acc, #eacc, effective accelerationism, kardashev, nuclear power, fusion, gigawatt, battery storage, data center power, small modular reactors, CANDU, IRENA, Starship and solar power. Kept on-topic posts (energy, compute, space, e/acc); skipped token promotion, off-topic posts and partisan political posts. Engagement is exactly what the API returned at fetch time.

X WATCH: 92 ANALYSED POSTS →