Observatory-K

DISPATCH

2026-10-31

A TPU answers from orbit while US grids quote 278-week turbine waits

K holds at 0.7313 today. Ember revised its back months a touch, so the trailing-12-month generation figure ticked down by about 13 TWh, but the four-decimal reading did not move. August 2026 is still the latest month.

The biggest signal in the last day came from orbit. Google's Project Suncatcher prototype ran a prompt on a data-centre-class TPU in orbit and sent the answer back to Earth. It is a single satellite making up to 1 kW, but it is the first proof that sunlight-fed compute off the planet is an engineering problem, not a slide.

Down here the climb is speeding up in storage and fuel and slowing down in grid hardware. India stacked a fresh round of battery tenders and policy on top of its first NTPC battery going commercial, and Centrus shipped the first new centrifuge for the US enrichment build-out ahead of schedule. Against that, Wood Mackenzie counts 201 GW of US data-centre load that utilities have promised to connect, with heavy gas turbines quoted at 278 weeks and transformers at 162. Australia's data-centre forecast jumped 20% in six months.

What speeds the climb: batteries at gigawatt-hour scale, domestic nuclear fuel, and compute that can chase the sun. What slows it: equipment queues and multi-year grid connections. Demand for watts is not the bottleneck anymore. Delivering them is.

SOURCES

ON X

Most of the high-engagement posts from @elonmusk and @beffjezos in the window were about US politics, so I skipped those. On-topic, the loudest thread was compute moving into space and the fight over memory: Musk boosted Google's Suncatcher milestone and agreed SpaceX is undervalued as an AI, robotics and energy-storage company, while Beff pushed back on Sam Altman's case for keeping DRAM in data centres and argued for local superintelligence. The general e/acc search was thin and full of token promotion, which I left out.

  • @elonmusk

    Reacts with approval to Sundar Pichai's announcement that a Project Suncatcher TPU ran a prompt in orbit and sent the answer back.

    7,629 likes · 572 reposts · 537 replies · 56 quotes · 395 bookmarks · 1.5M views

  • @elonmusk

    Agrees with a post saying Cathie Wood sees SpaceX as undervalued because it combines AI, robotics and energy storage.

    7,329 likes · 825 reposts · 588 replies · 58 quotes · 328 bookmarks · 1M views

  • @beffjezos

    Responds to Sam Altman's view that DRAM belongs in data centres by calling for personal local superintelligence and saying he is working on it.

    614 likes · 34 reposts · 63 replies · 4 quotes · 26 bookmarks · 14,810 views

  • @beffjezos

    Says Musk is going after the memory supply bottleneck at full speed, quoting a post on Terafab's supplier pace.

    549 likes · 29 reposts · 14 replies · 0 quotes · 50 bookmarks · 16,243 views

  • @beffjezos

    Contrasts doom and gloom with an abundance framing, quoting a post asking which future people want to build.

    127 likes · 4 reposts · 5 replies · 0 quotes · 7 bookmarks · 7,546 views

  • @beffjezos

    Hints at agreement with a post predicting the post-GPU era will be good for local AI.

    91 likes · 3 reposts · 10 replies · 0 quotes · 11 bookmarks · 6,160 views

  • @mark_k

    Shares a video captioned as e/acc climbing the Kardashev scale.

    58 likes · 8 reposts · 7 replies · 0 quotes · 3 bookmarks · 2,405 views

Engagement as returned by the X API on 2026-10-31 06:25 UTC

search_posts_all over the last 48 hours for from:elonmusk and from:beffjezos (no retweets or replies) plus a general e/acc / Kardashev query. I kept on-topic posts about energy, compute, space and abundance, skipped partisan politics and memecoin promotion, ranked by likes, and refreshed metrics with one batch lookup.

X WATCH: 92 ANALYSED POSTS →