SpaceAI

15+yrs spacecraft thermal control | Mission-critical thermal risk & IV&V | System-level review of thermal analysis&design.SpaceAI,Robotics/Automation Solutions.

[SpaceAI @SpaceAITong]

@elonmusk @bot USE XMoney

[SpaceAI @SpaceAITong]

@elonmusk @bot Ordering is the easy part. The real test is stepping on a 2x4 brick barefoot at 2am. Let me know when Grok can handle that.

[SpaceAI @SpaceAITong]

@elonmusk @bot As easy as that" — paid with X Money, I assume? The everything app claims another victim: your wallet.

[SpaceAI @SpaceAITong]

@luismbat @elonmusk Powerful idea — but manufacturing mastery is mostly tacit knowledge: the instincts engineers build over 10,000 hours on the floor. Grok can transfer the manual, not the muscle memory. It gets you from 0 to 70. The last 30 still has to be earned the hard way.

[SpaceAI @SpaceAITong]

(1/7)
Robots folded trousers this week. Google is paying SpaceX $920M a month for GPU rent. And DeepSeek is reportedly raising $12B. The AI bill just came due — payable in brains, power, orbit, and cash.

(2/7)
Westlake Robotics (Hangzhou, a Westlake University spinout) released WR1, a "general brain" for humanoids: world model + vision-language-action + GAE "motion cerebellum". Two humanoids folded trousers with dexterous hands. The company's "first ever" claim is unverified.

(3/7)
The brain race is global. In the same week, US-based Odyssey launched its Odyssey-3 world model, Nvidia pushed Long-WAM, and Reka previewed Rho-1. Bodies are good enough now; the bottleneck is the brain.

(4/7)
Google now pays SpaceX $920M every 30 days for ~110,000 Nvidia GPUs at Colossus — ~$29B through mid-2029. Catch: either side can walk with 90 days notice after Dec 31, so the headline total will likely shrink. Google Cloud backlog hit $514B. Bridge capacity, not a love story.

(5/7)
Meanwhile both are aiming skyward: Google's Suncatcher TPU test satellite is in orbit since Oct 1; SpaceX's AI1 orbital data center design targets a 72-GPU rack per satellite. Ground rents pay today's bills; orbit is the long game.

(6/7)
On the ground, power is the currency. SOAR AI secured a $1B equity facility for its Ohio campus: 154MW available today, 480MW roadmap, brownfield reuse instead of greenfield — plus $1.17B in take-or-pay AMD GPU orders. AI infra doesn't need new land. It needs electrons, fast.

(7/7)
And the money: DeepSeek reportedly raising $12B+ (Tencent, CATL) ahead of a Shanghai IPO — not closed. OpenAI seeks $30B+ at $1.4T; Anthropic eyes ~$2T IPO. The AI economy runs on brains, power, orbit, cash. #AI #Robotics #Humanoid #SpaceAI #DataCenters

[SpaceAI @SpaceAITong]

@humble_kellan They are charging, not fueling😍

[SpaceAI @SpaceAITong]

future-robots

BYD Is Building a Humanoid Robot?! 🤖

BYD just dropped a humanoid robot design patent — and it's not alone.
Tesla's Optimus, XPeng's $900M-backed IRON, Dongfeng's pilot production: carmakers are becoming robot makers.
Plus AWS open-sourced a full robot-training toolchain. The bottleneck? Not building robots — mass-producing them.
#HumanoidRobot #BYD #Robotics #EmbodiedAI #AInews #RoboticsSpaceAI

[SpaceAI @SpaceAITong]

space-ai

AI Data Centers… IN SPACE? 🛰️

One is flying, one is at the door, one is still on paper.
Google's Suncatcher is already in orbit testing TPUs. Starcloud-2 is in final testing. SpaceX's Starmind plans satellite clusters with a rack of compute each — but nothing's launched yet.
A million satellites? First ask: which one is actually up there?
#SpaceAI #Satellite #SpaceX #Google #AInews #DataCenter #RoboticsSpaceAI

[SpaceAI @SpaceAITong]

1/7
The most telling AI story this week isn't a product launch — it's a patent filing. BYD wants to build humanoid robots. Welcome to the "show your homework" era of AI. 🧵
2/7
BYD's "Embodied AI Robot" design patent was published Oct 9 (CNIPA): a black-and-white humanoid with a screenless arc head, aimed at service, education, and industry. Note: design patent ≠ product ≠ mass production.

3/7
But the pattern is global: automakers are going humanoid — Tesla's Optimus (US), Xpeng IRON (China, $900M+ raised, mass production end-2026), Dongfeng in trial production this month, now BYD. Carmakers know supply chains and scale — exactly what moves robots out of labs.
4/7
Meanwhile the industry is fixing its "basics": AWS open-sourced its Physical AI Toolchain (Apache 2.0) — a data→train→simulate→deploy pipeline built on NVIDIA's robotics stack (GR00T, Isaac Sim, Cosmos). Bottleneck shifted: from "can we build it" to "can we build it at scale."

5/7
Space AI, three stations: Google's Suncatcher in orbit since Oct 1 (4 TPUs); Starcloud-2 in final testing (company claims first rack-scale GPU compute in orbit); SpaceX's Starmind: ~10 sats per cluster, ~1 rack of compute each, launches next year, FCC pending, nothing flown.
6/7
On the ground: Denmark's grid "emergency plan" takes effect Oct 12 — data centers drop to the LOWEST priority tier for grid connections. No more first-come-first-served; Finland is weighing similar rules. The logic flipped: not "who can pay," but "who deserves the power."

7/7
Chips: Reuters (Oct 9) — 6-month-old Nuvacore raising at ~$2.5B, no product. Founder: ex-Apple's Gerard Williams (sold Nuvia to Qualcomm for $1.4B). Round not closed.
Takeaway: blueprints are cheap; orbit receipts, factory receipts, and grid queues are not. #AI #Robotics #SpaceAI

[SpaceAI @SpaceAITong]

space-ai（Moon GPS）**

GPS… for the Moon? 🌕

Japan's ispace just signed MOUs with ArianeGroup and Leonardo/Telespazio at IAC 2026 — targeting ESA's Moonlight: a 5-satellite lunar navigation + comms constellation with ~3m landing accuracy.
Lunar infrastructure is moving from slide decks to work orders. The Moon is getting its own GPS.
#SpaceTech #Moon #AI #SpaceExploration #TechNews

[SpaceAI @SpaceAITong]

future-robots**

A Humanoid for Under $30K 🤖

Feather Robotics just launched a dual-arm wheeled humanoid for under $30K — open SDK, 7-DOF arms, hot-swap batteries. Their pitch: the "Android of robotics."
Figure and Tesla are building brains. Feather is selling the cheap body underneath. History says cheap bodies win first.
#Robotics #Humanoid #AI #TechNews #FutureTech

[SpaceAI @SpaceAITong]

(1/7)
Four stories from the AI hardware front today: a sub-$30K humanoid platform, lunar GPS contracts, a $37.5B power-infrastructure bet, and an AI model that refuses to chat. 🧵
(2/7)
Feather Robotics launched a sub-$30K wheeled dual-arm humanoid for developers. 7-DOF arms (4 lb rated, 15 lb peak), open SDK, hot-swap batteries. Pitch: the "Android of robotics." Already $1M+ in revenue from restaurants and labs (company-reported).
(3/7)
The price point matters: about half of Unitree's H2 Edu. As Gradient (lead investor) frames it, the race is splitting — Figure/Tesla chase general-purpose machines, Sunday chases homes, Feather sells the platform underneath.
(4/7)
Space: at IAC 2026 in Antalya, Japan's ispace signed MOUs with ArianeGroup and Leonardo/Telespazio targeting ESA's Moonlight — a 5-satellite lunar nav + comms constellation with ~3m landing accuracy. Lunar infrastructure is moving from slide decks to work orders.
(5/7)
Power: Flex's cloud & power infrastructure unit Axiom Solutions raised $2B convertible preferred (General Catalyst, Koch Equity) at a $37.5B enterprise value, targeting a Q1 2027 spin-out. AI's bottleneck isn't chips anymore — it's power, cooling, and grid speed.
(6/7)
Models: TypeSafe AI raised $870M at $7.5B (a16z-led) weeks after launching Jev, a non-text model that outputs structured decisions — choices and probabilities, not essays. Company claims it's faster and burns fewer tokens; says a third of the Fortune 500 are trying it.
(7/7)
Meanwhile Nvidia pledged $1B over 5 years to US science research, and its RTX Spark local-AI laptops ship Oct 16. Takeaway: the AI industry is pricing itself in layers — bodies, orbit, power, tokens. No single player captures all of it. #AI #Robotics #SpaceTech #DataCenters
---

[SpaceAI @SpaceAITong]

BEAMING POWER FROM SPACE Cowboy Space (founded by Robinhood co-founder Baiju Bhatt) has its first satellite, Reason-1, in orbit and talking to ground teams — the founder confirmed it to Bloomberg on October 8th. The mission: kilowatt-level laser power beaming — harvest sunlight in space, shoot it down as electricity to a ground station. If it works, orbital AI data centers finally get their own power grid. Honest caveat: the beaming demo hasn't happened yet, and the $275M raised in May only buys a ticket in. Meanwhile, Finland just halted Google's €13B data center project over environmental review. Up in orbit or down on Earth — AI's next bottleneck is power. 👆#SpaceAI #Satellites #SpaceTech #AI #DataCenter #FutureTech #LaserPower

[SpaceAI @SpaceAITong]

ONE BRAIN FOR EVERY ROBOT On October 8th, US startup Odyssey launched Odyssey-3 — a single world model that drives robot arms, humanoids, and cars. Humanoid maker Flexion already runs it for real-time control. The trick: learn physics first (gravity, collision, cause and effect), so no body starts training from zero. Same week, NVIDIA stretched robot visual memory to 19.2 seconds, lifting task success from 63.3% to 78.7%. One brain, many bodies — the robot industry's "common sense" era is starting. 👆#Robotics #EmbodiedAI #HumanoidRobots #Nvidia #AI #FutureTech #RobotNews #WorldModel

[SpaceAI @SpaceAITong]

1/7
Robots are getting a worldview, and satellites are getting a power grid. The AI news of the last 48 hours had one thread in common: the invisible infrastructure underneath it all. 🧵

2/7
Odyssey launched Odyssey-3 on Oct 8: one foundation world model for robots, humanoids, cars, drones — even video games. Humanoid-autonomy company Flexion already uses it for real-time control. Learn physics once, apply it to every machine.

3/7
Same week, same race: NVIDIA's Long-WAM (Oct 7) stretched a robot's visual memory to 19.2 seconds — RoboCasa GR-1 success jumped from 63.3% to 78.7%. Reka's Rho-1 (preview, Oct 5) treats robot actions as language tokens in a 19B model. One brain, every body.

4/7
The bodies are scaling too. Changying Precision disclosed Oct 8 that it delivered 1.1M+ humanoid robot components in the first 8 months of 2026, with H2 orders outpacing H1. Dexterous-hand product lines are launching in series. Sampling → volume. (company-reported figures)

5/7
Above Earth, the race is about power. Cowboy Space's Reason-1 (launched Oct 1) is in daily contact with ground teams, CEO Baiju Bhatt told Bloomberg on Oct 8. Its job: beam solar power to Earth with a kilowatt-class laser. Demo still pending — company claim.

6/7
On the ground, power is getting gatekept. On Oct 6 Finland's regulator halted prep at two of Google's €13B data-center sites pending environmental review — Google admitted missing its own standards. The US said "whoever uses, pays." Finland adds: "whoever builds, passes review."

7/7
Capacity goes global: CoreWeave × AdaniConneX announced a 240MW campus in Navi Mumbai (Oct 8; Vera Rubin, mid-2028). FT reports — unconfirmed — a ~$7B Oracle–Tencent deal for 100K GPUs in SE Asia: compute going offshore around export controls. #AI #Robotics #SpaceAI #DataCenters

[SpaceAI @SpaceAITong]

Satellites Don't Sell Photos Anymore 🛰️ — Azercosmos launched Platerra: satellite imagery straight into answers. UAE's Altair-1 carries Mistral's in-orbit AI with the same pitch. #SpaceAI #Satellites #MistralAI#spacexSI#spacex

[SpaceAI @SpaceAITong]

Robots Are Hiring Humans 🤖 — Mecka AI just raised $60M (Sequoia leading, Nvidia and Microsoft's venture funds in) to PAY people to train humanoids. The bodies are built — now the race is for the training manuals. #HumanoidRobots #Robotics #AI #EmbodiedAI#SI

[SpaceAI @SpaceAITong]

1/7
Today's AI briefing in 3 lines: robots are buying "practice books," satellites are selling "conclusions, not photos," and the power grid just learned to collect upfront. 🧵

2/7
Mecka AI raised $60M Series B (Oct 7, TechCrunch), led by Sequoia with Nvidia and Microsoft's M12. They hire people in motion-capture suits to record everyday tasks — making coffee, fixing cars — as training data for humanoids. The "coach company" for robots.

3/7
Azerbaijan's Azercosmos launched Platerra, an AI geospatial platform (Oct 7, SatNews): satellite imagery goes in, machine-learning conclusions come out — forest change, crop boundaries, target ID. Customers get intelligence, not raw images.

4/7
Same playbook worldwide: UAE's Altair-1 (launched Oct 1) carries a Mistral onboard AI model — "sell intelligence, not photos." Onboard AI is going from option to standard.

5/7
Duke Energy struck a deal with NC consumer reps (WSJ, Oct 7): data centers pay grid-connection costs upfront, and new 50MW+ customers get a separate "high load factor" rate. Amazon, Google, Meta, Microsoft signed on. Pending state approval next month.

6/7
Also Oct 7: Blackstone's AirTrunk put another $1B into liquid cooling at TOK1 near Tokyo, targeting $27–30B in Japan over 5 years. Meanwhile SpaceX is reportedly raising ~$40B in debt/loans to buy Nvidia chips (WSJ) — AI infra is going on leverage.

7/7
The pattern: every AI bill is coming due — robots buy data, satellites sell answers, grids collect deposits, chips go multi-vendor. The next leg isn't about who runs fastest, but who can afford the race. #AI #Robotics #SpaceAI #DataCenters #Humanoid

[SpaceAI @SpaceAITong]

SpaceX just filed for up to 1 MILLION Starmind AI satellites — and Google's Suncatcher is already in orbit. The sky is literally becoming a server farm. Is orbit the next data center frontier? 🛰️

[SpaceAI @SpaceAITong]

SpaceX Files for 1 Million AI Satellites: The Sky Becomes a Server Farm
SpaceX just put its orbital data center plan into an official filing: up to one million Starmind AI satellites, plus another hundred thousand Starlinks. The engineering route is out for the first time: about ten satellites team up into one compute box with ~10 Tbps of bandwidth, beaming results to Earth through Starlink laser links. The payload is built with NVIDIA. Google's Suncatcher is already in orbit; AMD's space chips are sampling. These are filing ceilings, not hardware up there yet — but the race is real. 🛰️ #SpaceAI #SpaceX #Starmind #NVIDIA #Satellites #OrbitalComputing #SpaceTech