S5 Ep. 7 - Why future of computing might not be on earth

Impossibile aggiungere al carrello

Puoi avere soltanto 50 titoli nel carrello per il checkout.

Riprova più tardi

Rimozione dalla Lista desideri non riuscita.

Riprova più tardi

Non è stato possibile aggiungere il titolo alla Libreria

Per favore riprova

Non è stato possibile seguire il Podcast

Per favore riprova

Esecuzione del comando Non seguire più non riuscita

S5 Ep. 7 - Why future of computing might not be on earth

Ascolta gratuitamente

Vedi i dettagli del titolo

Offerte di stagione | 0,99 €/mese per i primi 3 mesi

A seguire 9,99 €/mese – si applicano condizioni. Puoi disdire mensilmente.

Elon Musk just merged xAI into SpaceX with a plan that sounds absurd until you run the numbers: build data centers in orbit.Not small experiments but industrial-scale computing infrastructure floating 340 miles above Earth (same distance of NASAs telescope Hubble), powered by solar arrays, cooled by the vacuum of space, training AI models that would potentially bankrupt terrestrial facilities.The immediate reaction is skepticism. Rockets are expensive. Space is hard. Servers belong in buildings, not orbit.But strip away the science fiction framing and what remains is a serious economic calculation that every executive should understand: sometimes the cost of working around a constraint exceeds the cost of eliminating it entirely.Nature.com - The development of carbon-neutral data centers in spaceThe constraint no one can optimize AI training has hit a wall. Not a technical wall. A physics wall. Training GPT-4 required approximately 50 gigawatt-hours of electricity. The next generation requires exponentially more. Frontier models now consume energy equivalent to small countries (es. Belgium), running for weeks or months continuously.Data centers already eat 1-2% of global electricity. AI is accelerating that dramatically. And here's the killer: in a typical hyperscale facility, cooling accounts for 38-40% of total energy consumption. For every watt your servers burn doing computation, you spend nearly another watt just preventing thermal failure.This isn't inefficiency but physics. Dense computing generates massive heat. Remove it continuously or the hardware dies.The industry has optimized relentlessly. Better airflow. Liquid cooling. Chip-level thermal management. But optimization has limits, and those limits are approaching faster than most realize. You can only pack so much compute into a building before cooling becomes impossible regardless of technology.Meanwhile, AI's appetite keeps growing. Meta, Google, Microsoft, and others are each planning to spend over $60 billion annually on AI infrastructure. Much of that goes to power and cooling.This is the constraint that changes everything. Not because it's insurmountable today, but because it's exponentially worsening and the traditional solutions don't scale.Haysfluidcontrols.comMath that makes orbit interestingMoving data centers to space sounds expensive until you compare it to the alternative.A 100-megawatt terrestrial AI facility costs $1-1.5 billion to build. Over ten years it consumes roughly 8,760 gigawatt-hours of electricity. At $60 per megawatt-hour, that's $525 million just for power. You spend 38-40% of that, approximately $200 million, on cooling alone. Add construction, land, taxes, maintenance and you're approaching $2 billion total cost of ownership.Now consider orbital economics: Historically (up to year 2000s), this calculation was laughable. Launch costs of $10,000-20,000 per kilogram made space infrastructure prohibitively expensive for anything except satellites and scientific missions.SpaceX's Starship changes this fundamentally. Current projections target $100 per kilogram to low Earth orbit, potentially as low as $10 per kilogram at full operational scale. That's not incremental improvement. It's two to three orders of magnitude cheaper.A server rack weighs 1,000-2,000 kilograms. At $100 per kilogram, launching one costs $100,000-200,000. At $10 per kilogram, it's $10,000-20,000. For comparison, the servers themselves often cost more than their launch at these prices.But weight adds up. A meaningful orbital data center requires thousands of racks, solar arrays, thermal radiators, communication equipment, structural elements. Even at optimistic weights, you're launching hundreds of thousands of kilograms.Here's where the calculation gets interesting. At $10-100 per kilogram, putting 100,000 kilograms of infrastructure in orbit costs $1-10 million for launch. The equipment itself might cost $50-100 million. Total upfront: $51-110 million.Made with Claude ProOperational costs nearly disappear. Power is free once solar arrays are deployed. Cooling requires zero energy because thermal radiation to space is passive. No property taxes. No grid fees. No atmospheric corrosion. Maintenance becomes the primary ongoing expense.Compare that to terrestrial alternatives where you're spending $200 million on cooling energy alone over a decade, and the economics start shifting.Problem 1: Maintenance The most legitimate objection to orbital data centers isn't launch costs or power but maintenance.Terrestrial data centers have technicians on-site twenty-four hours a day. When a server fails, someone swaps it within hours. When a cooling system malfunctions, engineers fix it immediately. When software needs updating, IT teams access the hardware directly. This operational flexibility is taken for granted, but it's fundamental to how data centers work.In orbit, none of this exists. You cannot send a technician to replace a failed drive. You ...

Ancora nessuna recensione