How to connect space-based AI clusters to Earth: Aalyria's solution

Space-based data centers are no longer a thought experiment. Huge congratulations to our friends at Google on Project Suncatcher — a serious, systems-first exploration of on-orbit compute. 👏 Let me break down the requirements of deploying such a system and how I think Aalyria can be a valuable asset in making it happen. How much ground ↔︎ space I/O will an orbital cluster need? Recent measurements from Sandvine’s put total global Internet traffic at roughly 33 exabytes per day, with a small set of hyperscaler platforms (including Google/YouTube, Amazon, Meta, Microsoft) together accounting for around half of that volume. Spreading this multi-exabyte load across on the order of a hundred-plus Google-scale facilities implies that a single modern data center is engineered for multi-terabit-per-second external ingress+egress— with tens of petabytes per day moving in and out of the facility. In Project Suncatcher, Bloom et al. describe an 81-satellite TPU cluster whose performance is “roughly comparable to a terrestrial datacenter,” linked to Earth via optical ISLs and dedicated feeder links to ground; so, if we assume external I/O scales with compute, a conservative estimate is that each Suncatcher-class on-orbit cluster would require on the order of ~1–5 Tb/s of aggregate Earth–space feeder-link capacity. Where Aalyria fits: Tightbeam (optical ground terminals): designed to make those feeder links possible. Our early units support 100-400Gbps full duplex per terminal (range & conditions dependent) at 1550 nm. You would deploy several per site — and across multiple sites, for weather/site diversity. That’s how you support Tbps-class clusters on the feeder-link side. Spacetime (temporospatial SDN): once you run optical feeder links through air, weather will impact the network. Spacetime proactively retasks steerable terminals — on the ground and in space. So if Spacetime forecasts clouds obscuring one site, we’ll pre-switch to a clear site and reconcile the on-orbit inter-satellite link graph, because a different ground look angle often means a different/adjacent satellite should carry the hop. Spacetime is built to optimize the link topology, spectrum resource allocation, and path computation across L1/L2/L3 concurrently, anticipating fades and evolving paths in sub-second timeframes (we’ve demonstrated sending updates to reconstitute a new topology after a satellite failure in less than 200 ms). This is the control plane you need when the physical fabric itself is dynamic. 💡Put simply: If Suncatcher makes the compute and power story compelling, Tightbeam and Spacetime can make the networking story real. Early systems can start with a few hundred Gbps of optical feeder capacity and scale out to multi-site, multi-terabit aggregates — while Spacetime keeps the whole thing stitched together as weather and orbital geometry evolve. https://lnkd.in/et2pnNei

Ars Technica writes "Meet Project Suncatcher, Google’s plan to put AI data centers in space - Google is already zapping TPUs with radiation to get ready." https://lnkd.in/eXpcYR4a #google #projectsuncatcher #datacenter #inspace #tpus

Compute in space isn’t new. I wish journalists would dig a little deeper; you’d think LLMs could help. Every time someone announces compute in orbit, the headlines make it sound brand new. This week it’s Google, TPUs in space! as if no one had ever looked up and had that thought before. Google’s Project Suncatcher — a “moonshot” plan for solar powered AI data centers in orbit (The Verge): https://lnkd.in/gnNgcdFf But of course, this isn’t new. In 2018 I was asked to speak at Space 2.0 in San Jose about AWS in Space, 8 years after GPUs came to EC2 in 2010. At the time, I was the Geospatial Lead on the AWS Solutions Architecture Specialist team. When they first asked me, my reaction was something like: Why do they want me to talk about compute in space when most people are still struggling to migrate to the cloud? Aren’t we getting ahead of ourselves? But the premise made sense: Do the math where the data is born. Satellites generate oceans of data, why downlink terabytes when you can process in orbit and only beam down what you actually need? Most of what we do in space supports Earth observation: defense, weather, climate, mapping, agriculture, disaster response. EO is why space matters day to day. AWS was already extending its network: • Regions → Availability Zones → Local Zones → Edge with AWS Snow devices → Orbit • Ground Station (announced at re:Invent 2018) providing connectivity between orbiting assets and AWS Regions • Amazon S3 as the landing zone for large earth observation datasets (e.g., AWS Open Data) • Amazon Kuiper (disclosed in 2019) as an emerging LEO communications layer integrating with AWS services In 2022, the AWS Aerospace & Satellite team flew a Snowcone device for an on orbit demo (ISS), a small edge compute unit designed to run AWS services at the edge: https://lnkd.in/gvEn389f Looking back, there was no magic. No moonshot branding. Just basic distributed systems thinking, extended a little farther from home in pragmatic, stepwise fashion. Very Amazonian. In the long run, the winner in space computing won’t be the one with the biggest orbital AI cluster, but the one that connects the full pipeline from space to Earth and back, from sensor to storage to insight, seamlessly tied to enterprise data and built on economics that scale across industries. So yes, Google’s Project Suncatcher concept is ambitious. It’s good to see more companies thinking about compute beyond Earth and keeping the thermal footprint off planet. But these ideas aren’t new; they’ve been part of the cloud conversation for years. #CloudComputing #SpaceTech #AWS #AIinSpace #Geospatial #EarthObservation #DistributedSystems #Aerospace #GPUComputing

Google’s Project Suncatcher Aims to Build TPU Powered AI Data Centers in Space #Google #AI #ProjectSuncatcher #SpaceTech #DataCenter #Satellites #TPU #Moonshot #CloudComputing #FutureofAI #Alphabet #GoogleAI #SpaceEconomy https://lnkd.in/eYtFcyqr

In the summer of 2014, I was interviewing with a major tech company. During a phone interview for a system design round, a senior architect asked me a question I still remember: “If we were tasked with setting up data centers in space, how would we design them? What key considerations should we follow?” At the time, my family and I had just returned from a trip to Orlando where we had spent a day visiting the Kennedy Space Center. Drawing on what I had learned that day about life in space (the thermal constraints, radiation risks, energy generation) I sketched out a design grounded in reality. I passed the round. I remember thinking: They must already be thinking about data centers in orbit. Fast forward to this week – and now Google is doing just that. Their newly announced Project Suncatcher envisions launching constellations of solar-powered satellites, each carrying Google’s TPU (Tensor Processing Unit) AI chips. These satellites would orbit in a dawn-dusk, sun-synchronous path, keeping them bathed in almost constant sunlight. Why does this matter? Because it’s an audacious response to one of AI’s biggest sustainability challenges: how to power compute at scale without burning out Earth’s energy system. On Earth, data centers consume massive amounts of electricity and water – and building more of them carries environmental and social costs. Google’s vision: use space to tap into “near-unlimited” solar energy, reducing strain on terrestrial infrastructure. Reading about Project Suncatcher brought me right back to that interview moment. It’s wild to think that what once felt like a purely hypothetical question may actually be coming true. https://lnkd.in/emgv3xdK

🚀 Big news from Google — in an article by Reed Albergotti at Semafor ( https://lnkd.in/gSsUeh6u ), the company outlines a wildly ambitious concept: “Project Suncatcher”, where solar-powered data centers are placed in orbit. If you've ever read Delta-V or Critical Mass by Daniel Suarez (I'm a huge fan of all of his books!), this might sound somewhat familiar... Here are the standout points: 👉 Google says the surge in demand for AI compute means Earth-based power and data centers may not scale efficiently. 👉 Their idea: launch AI chips (TPUs) into low-Earth orbit, in constant sunlight, enabling continuous solar power — up to 8× the annual solar energy compared to ground panels at mid-latitudes. 👉 The satellites would form a tightly-linked constellation (hundreds of meters apart) with high-bandwidth wireless/optical links between them. 👉 First test: two satellites, each carrying four TPUs, planned for 2027 with Planet Labs. If launch costs drop (from ~$1,500/kg to ~$200/kg by ~2035), this could become cost-competitive with terrestrial data centers. 👉 Challenges remain of course: radiation exposure in space, chip lifespan, the carbon/launch-emission footprint vs. earth-based centers. A recent EU-funded study flagged the need for reusable low-emission rockets (<370 kg CO₂/kg payload) for the concept to make climate sense. 🔍 “If things keep going down the path where we keep having more uses for AI and we keep wanting more energy … this has tremendous potential to scale,” says Google researcher Travis Beals. Personally I love watching companies taking the moonshots to solve big problems. Whether it succeeds or fails, it'll be a fun journey with plenty to learn! #AI #DataCenters #Infrastructure #Innovation #Sustainability #Google

Google's Moonshot Plan: AI Data Centers in Space Google has an ambitious "moonshot" plan to develop AI-driven data centers in space using constellations of solar-powered satellites, enabling nearly continuous solar-powered operation for its TPU chip infrastructure. In a bold move, Google is venturing beyond traditional terrestrial data centers by conceptualizing the deployment of AI-driven data centers in outer space. The visionary plan hinges on utilizing solar-powered satellites, which can harness solar energy almost continuously, thus providing a consistent power supply to the company's TPU chip infrastructure. Google's initiative involves creating constellations of satellites equipped with advanced AI capabilities. These satellites will serve dual purposes—act as data centers and harness solar power, thereby mitigating the typical limitations of space-based energy generation. While the prospects are exciting, building data centers in space is fraught with challenges. For instance, the cost of launching and maintaining satellites, potential space debris issues, and ensuring robust cybersecurity measures to protect data against hacking or cosmic interference are significant hurdles. As Google explores this new frontier, the impacts are likely to resonate throughout the tech industry, potentially prompting other companies to adopt space-based data solutions. It also opens up avenues for innovation in satellite technology and space exploration. In another dimension, Google's endeavor could contribute to the network infrastructure essential for future space colonies, potentially positioning AI as an integral part of mankind's extraterrestrial expansion. As scientific curiosity aligns with technological advancement, the sky is no longer the limit. #CurrentTrendsInTechnology Source : The Verge

This could redefine the physical boundaries of computation. Google just unveiled Project Suncatcher, a moonshot initiative to launch AI data centers into space. Yes—actual orbital infrastructure powered by uninterrupted solar energy and high-speed satellite networking. It is an interesting solution to augment the 4,165 active data centers in the United States. Many of these data centers tend to be concentrated in areas such as Loudoun County, VA; Maricopa County, AZ; Santa Clara County, CA; and Cook County, IL (just to name a few). These strain local grids and ecosystems. Moving compute off-planet could reduce land use, water consumption, and community opposition. It will be interesting to see how Google will solve the following risks; > Space debris and orbital congestion > Latency and bandwidth constraints > Cybersecurity and jurisdictional ambiguity > High costs and unproven long-term viability Would love to hear your thoughts: Is space the next frontier for cloud infrastructure, or just a high-altitude experiment? #AI #CloudComputing #Google #ProjectSuncatcher #DataCenters #SpaceTech #Sustainability #Innovation #Moonshot https://lnkd.in/gb7yRzaN

Datacenters running LLMs in space?? It sounds me like Sci-Fi story, but here's what I found... Ever thought of data centers orbiting in space? As we see the exponential growth in LLMs usage, two things drive accuracy: huge datasets and massive compute power. Big initiatives are underway to deploy new data centers - but I just read about a project from StartCloud: the launch of extraterrestrial data centers. Then I researched about it why pushing data centers to orbit? In space, you tap into virtually unlimited, low-cost renewable energy. After the one-time launch cost, Starcloud claims a 10x reduction in CO2 emissions compared to powering terrestrial data centers, unlocking affordable AI at scale. Okay what's up for the business? Starcloud estimates that energy costs in orbit will drop by a factor of ten, making LLMs dramatically cheaper and opening the door to widespread LLM adoption across industries, which is good. I still have a lot of questions like what happens if an entire data center shuts down, or how they plan to decommission specific components. I've been exploring these topics further on my Perplexity pages. Here’s the link: https://lnkd.in/gt4K4ftV #AI #LLM #ZeroCarbon

Google just announced Project Suncatcher — a plan to launch solar-powered AI data centers into orbit by 2027. The idea is to take advantage of constant sunlight in space to power advanced AI systems while reducing the strain that massive data centers are putting on Earth’s power grids, land, and water supplies. These orbital servers would use Google’s TPUs (the chips that power AI models) and communicate through high-speed laser links capable of moving data at incredible rates. It’s an ambitious concept — but not without hurdles: Keeping hardware cool and protected from radiation Maintaining strong, low-latency connections between space and Earth Managing launch costs and avoiding orbital debris Figuring out how global data laws apply to servers floating above the planet Why it matters: Data centers now consume as much power as small cities, making it a real challenge to find sustainable ways to expand computing capacity. This project is Google’s attempt to rethink the problem entirely — by moving the cloud beyond Earth. It’s early, experimental, and far from reality, but it could mark the start of a new era in how we build and power technology. #Google #AI #DataCenters #CloudComputing #Sustainability #Innovation #FutureTech https://lnkd.in/eaPvnQHX