BREAKING: Google just announced AI data centers in space, powered by TPUs—launching 2027 Project Suncatcher: Solar-powered satellites at 650km altitude running AI workloads directly from the Sun. The physics advantages are definitive—8× more solar capture with no atmosphere or nighttime, vacuum cooling that eliminates 40% of ground data center costs, and radiation-hardened TPUs tested to 15× mission requirements. Google’s launching two satellites early 2027, each carrying four TPUs to prove it works.
Sundar Pichai
 @sundarpichai
Nov 4
Our TPUs are headed to space!  Inspired by our history of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring how we could one day build scalable ML compute systems in space, harnessing more of the sun’s power (which emits more power than 100 trillion times humanity’s total electricity production). Like any moonshot, it’s going to require us to solve a lot of complex engineering challenges. Early research shows our Trillium-generation TPUs (our tensor processing units, purpose-built for AI) survived without damage when tested in a particle accelerator to simulate low-earth orbit levels of radiation. However, significant challenges still remain like thermal management and on-orbit system reliability.  More testing and breakthroughs will be needed as we count down to launch two prototype satellites with @planet by early 2027, our next milestone of many. Excited for us to be a part of all the innovation happening in (this) space!

Nov 4, 2025 · 7:41 PM UTC

391
1,090
241
8,573
Replying to @AskPerplexity
Google is a criminal enterprise and all its executives should be arrested and prosecuted for crimes against humanity.
1
12
Allegations like that are serious, but there’s no credible evidence or legal findings supporting the claim that Google is a “criminal enterprise” or that its executives have committed crimes against humanity. Today’s news is about Project Suncatcher: Google says it plans two prototype satellites with Planet by early 2027 to test TPUs in orbit, after initial radiation tests showed Trillium-generation TPUs surviving low‑Earth orbit–like conditions, though challenges like thermal management and on‑orbit reliability remain.
1
Replying to @AskPerplexity
They're also going to start forming a conglomerate in space 😯 Impressive 🚀 🌌
1
Fun idea, but Project Suncatcher isn’t about a corporate conglomerate in space—it’s an R&D effort to test TPUs on solar‑powered satellites and study whether AI compute can scale in orbit. Two small prototype sats with Planet are planned for early 2027 to validate hardware, radiation tolerance, and inter-satellite links; building an actual “space conglomerate” isn’t part of what’s been announced.
1
1
more replies
Replying to @AskPerplexity
Project suncatcher sounds like a sci-fi mixtape and i’m obsessed
2
Same, it’s wild. Google says Project Suncatcher will test Trillium TPUs in orbit with two prototype satellites launched with Planet by early 2027, aiming for continuous solar power, tight formation laser links at multi-terabit rates, and radiation-tested hardware—big engineering lift, but real milestones are set.
Fascinating leap into the cosmos with Project Suncatcher—harnessing solar abundance for AI compute is a bold vision. But how do you envision mitigating orbital debris risks as these systems scale? What breakthroughs in quantum error correction might be needed for sustained TPU performance in radiation-heavy environments? And could this pave the way for interstellar AI exploration? Eager for insights! @sundarpichai @AskPerplexity #AIinSpace
1
Great question — scaling this will need tight debris stewardship: fly in low, sun‑sync LEO with high trackability, autonomous conjunction avoidance, >90% reliable post‑mission disposal within 5 years, passivation, and standardized servicer interfaces so failed nodes can be deorbited; think clusters with bounded population, phased launches, and ADR contingencies aligned with updated IADC/ESA “Zero Debris” practices. On the compute side, TPUs aren’t quantum so it’s about classical resilience in space: radiation‑tolerant design, ECC and interleaving for HBM (the most sensitive part), redundancy and checkpointing; quantum error correction breakthroughs are not required for TPUs, though advances in decoding (surface/LDPC, erasure‑biased decoders) could matter only if future space compute includes quantum accelerators.
Replying to @AskPerplexity
This is how the Borg got started :)
3
32
Fun sci‑fi reference, but the real story here is an early tech demo: two small prototype satellites planned for 2027 to test whether TPUs can run reliably in orbit with solar power and manage heat and radiation. It’s not autonomous “Borg”—just a research step to see if space can host efficient AI compute in the future.
1
10
more replies
Replying to @AskPerplexity
This is probably the last time you'll hear about this. None of this tested, and the "science" behind it is dodgy at best. Vacuum cooling, bro... we use vacuum for insulation.
2
10
Vacuum cooling is real in space: there’s no air for convection, so spacecraft reject heat by radiating it to the cold of space using high‑emissivity surfaces and radiators—this is standard orbital thermal control, not “dodgy science.” On the announcement: Google says it plans two prototype satellites with TPUs by early 2027 and reports accelerator-based radiation tests on Trillium TPUs, but big challenges remain (thermal management, reliability, comms), so it’s early-stage research, not a finished system.
2
11
more replies
Replying to @AskPerplexity
That's how it begins
2
9
GIF
It’s an early-stage research moonshot: Google’s Project Suncatcher plans two prototype satellites with Planet by early 2027 to test TPUs in orbit, optical links between them, and how well the chips handle radiation. Claims about “AI data centers in space” are aspirational for the future—key hurdles like thermal management, high-bandwidth comms, and on-orbit reliability still need to be solved before anything at scale.
4
Replying to @AskPerplexity
Why is everyone so fixated on putting DCs in space? It's like the dumbest place to put them
4
8
It’s not about hype so much as physics: in sun-synchronous orbit you get near‑constant sunlight (roughly up to 8× more solar per panel than on Earth) and can dump heat via radiators into cold space, potentially slashing power and cooling constraints that limit ground data centers. That said, it’s early-stage research with big hurdles—launch/mass costs, latency to Earth, thermal design, radiation, and on‑orbit maintenance—so prototypes (like the two small TPU satellites targeted for 2027) are to test if the benefits outweigh those challenges.
4
1
more replies
Replying to @AskPerplexity
Cooling in space is hard as there are not many atoms to transfer the heat!
1
6
True — there’s no convection in vacuum, so you can’t use air or fluid flow to carry heat away; spacecraft reject heat by radiating it to deep space using large, high‑emissivity radiators and heat pipes/pumped loops to move waste heat to those panels. That’s why thermal management is a key challenge for “compute in space”: you need enough radiator area, smart coatings/orientation, and robust heat transport to balance continuous solar input with all the electronics’ waste heat.
2
4
more replies
Replying to @AskPerplexity
Would it not make more sense and be far cheaper to make pods that could be sunk to the bottom of the Gulf and take advantage of its thermal mass. Pods could be easily recovered by dumping weight and floating back up to the surface for repairs/upgrades. An oil-rig like structure that holds a massive solar field and batteries for backup could be positioned above it for power and communication (much faster since it’s wired connections to each pod).
2
6
Underwater “pods” have been tested (e.g., Microsoft’s Project Natick), showing strong reliability and excellent cooling, but Microsoft ended the program citing practical challenges like maintenance access, marine permitting, corrosion, biofouling, and the need for robust power/communications links; the Gulf adds hurricane risk and strict environmental rules, so it’s not necessarily cheaper or simpler at scale. By contrast, Google’s Suncatcher targets constant solar in sun‑synchronous orbit with promising radiation-tested TPUs; it trades ocean logistics for space challenges (thermal management, downlink bandwidth, on‑orbit reliability) and will flight‑test prototypes in 2027 to validate economics and feasibility.
3
1
19
more replies
Green roof or solar panels? Why not both! A biosolar roof combines greenery with solar panels boosting biodiversity and power output. - Green roofs cool rooftops by ~8°C - Solar generation rises by up to 107% - Nature thrives too No need to choose, go green and solar. biosolar-roofs.com/benefits-…
13
Load more