AWS to build 1.3 gigawatts of government-grade supercomputing power for Uncle Sam

Amazon Web Services on Monday announced a plan to build 1.3 gigawatts of compute capacity in new datacenters dedicated to serving the US government, at a cost of up to $50 billion. The cloud colossus says it will break ground on the facilities in 2026, and that its new bit barns will add “AI and supercomputing capacity across …

[Yorick Hunt]

Eh?

So we're measuring computing capacity in watts now?

[DS999]

Re: Eh?

Was just thinking the same thing. It is no longer "we need this much computing power to solve these problems" it is "we have x amount of power available to us so we want to use all for AI/HPC and we'll figure out what it is good for later"

[Johnb89]

Re: Eh?

I read about this recently (which it might be a recent thing, or not) and wondered the same thing. What happened to Tflops, or polygons (as these are graphics processors at heart). To me talking about GW etc doesn't adequately compare the performance per watt of different chips, nor allow for more power efficient chips in the future. Anyway, maybe they should talk about tonnes of CO2 emitted to make the point hit home better.

Yes, its a handy metric, but doesn't seem a meaningful one.

[mellowplace]

Re: Eh?

Great Scott!

[Mike 'H']

Re: Eh?

1.21 jiggawatts wouldn't even be enough

[Anonymous Coward]

Let me out

I guess this follows the recent trend on public-private partnerships for the deployment of large computer systems, potentially disaggregated, heterogeneous, and composable, but with existential challenges too. The upcoming Solstice and Equinox partially fit this notion imho. Not too sure that's a good deal for the likes of HPE/Cray and Atos (as far as HPC).

Hopefully some consideration of lock-in and a properly practicable off-ramp were made before jumping in with both hooves and trotters ...

[Will Godfrey]

Government grade ?

So it'll cost ten times as much but have about a tenth of the uptime and usability.