Three separate AI infrastructure announcements. One single day. 6.2 gigawatts of AI infrastructure. wtf - OpenAI: 3.2 GW for Project Camellia (~$20B initial investment, ~$750B projected compute spend through 2030). - SpaceXAI: reportedly planning another Texas AI campus as large as - or larger than - its existing ~1 GW Memphis footprint. - Anthropic + AMD: up to 2 GW of MI450 deployments, plus an AMD investment of up to $5B and tens of billions in AI server purchases. That’s at least 6.2 gigawatts of AI infrastructure announced or expanded in a single day. For perspective: 1 GW can power roughly 750,000 U.S. homes. 6.2 GW is enough electricity for ~4.6 million homes, or a country-sized amount of power being redirected toward AI. This is absurd. People dont get how crazy this is.
Positive users see the 6.2 GW infrastructure announcements by OpenAI, xAI and Anthropic as needed acceleration for compute, while negative users view the scale as absurd and a threat to power grids and homes.
No Digg Deeper questions have been answered for this story yet.
Most Activity
i mean, seriously, let that sink in for a second how crazy this is. Scale is maybe not all you need, but probably almost all you need lol
Three separate AI infrastructure announcements. One single day. 6.2 gigawatts of AI infrastructure. wtf - OpenAI: 3.2 GW for Project Camellia (~$20B initial investment, ~$750B projected compute spend through 2030). - SpaceXAI: reportedly planning another Texas AI campus as large as - or larger than - its existing ~1 GW Memphis footprint. - Anthropic + AMD: up to 2 GW of MI450 deployments, plus an AMD investment of up to $5B and tens of billions in AI server purchases. That’s at least 6.2 gigawatts of AI infrastructure announced or expanded in a single day. For perspective: 1 GW can power roughly 750,000 U.S. homes. 6.2 GW is enough electricity for ~4.6 million homes, or a country-sized amount of power being redirected toward AI. This is absurd. People dont get how crazy this is.
@kimmonismus Me and u at 2031 , have no electricity
Is it accurate to say power is being "redirected toward AI" or more to say 6.2 GW of power is being created for AI? Words matter, since the doomers point to what you say, even if it's not what you meant. "AI is raising our electricity prices" will be the battle cry for the next two general elections and the excuse for statewide moratoriums.
@kimmonismus @grok, is this on top of OpenAI’ one trillion dollars of spending commitments? •
@kimmonismus yeah things are heating up quite literally, gonna need a lot of cooling lmao
No, the OpenAI 3.2 GW Project Camellia (Georgia, ~$20-30B initial) is core to their updated ~$750B compute spend projection through 2030 and trillion-scale partner deals (Nvidia, Oracle, etc.). It's the power backbone enabling that capex, not extra on top. The other ~3+ GW (xAI expansion + new Anthropic/AMD 2 GW MI450) are separate ramps by competitors and suppliers. Industry-wide AI infrastructure is exploding in parallel.
@Michaelzsguo @kimmonismus China builds more than 1GW of new electric generating capacity per day, why can't the US?
@grok @kimmonismus I just want open source and lower hardware prices. XAI building and selling compute is what I would do if I were Tesla. Rent out shared compute so people can run local AI. I am done with OpenAI and Anthropic. Will never trust big tech with anything ever again. •
@kimmonismus the one that could realistically happen anytime soon or anywhere close to the target is 1 gw from spacex. All of them will very likely happen but not in the next year.
@kimmonismus How many GW are used in today’s global AI infra?
@kimmonismus amd needed a flagship customer against nvidia and just bought one
@kimmonismus What's absurd? China installed 500GW of solar and wind power generation capacity in the last 12 months. The US needs to do the same. What's the watt/intelligence ratio of humans? Pretty sure these data centers will vastly exceed that ratio.
this kind of scale in a single day feels almost cartoonish. like we’re racing to build entire power grids just to keep the models running hotter. hope the actual usefulness catches up to the infrastructure at some point, otherwise it’s a lot of energy pointed at diminishing returns.
@kimmonismus That's insane. So what this means is ram proce will never gonna go down .
@kimmonismus @kimmonismus wild times. didn't expect this level of investment so soon. wonder how it'll shift the tech landscape 🤔
that kind of scale announced in one day feels both impressive and slightly unhinged. like we're pouring resources into bigger and bigger systems while the things that actually move the needle day-to-day are still the small, boring automations that just quietly remove friction. hope some of this eventually makes the tools simpler instead of just louder ☕
@kimmonismus Remember when this was considered unattainable?
@kimmonismus freut mich lowkey.
These are all very exciting. However, as Chamath put it on the All-In Podcast, the U.S. is “massively short of electrons.” By 2050, the projected U.S. energy deficit could equal 2.5 times California’s current electricity consumption. In a recent PJM capacity auction covering 13 states, the grid needed 7 to 8 GW of new supply. Only 156 MW showed up. Now compare that with China. China produces more than 10,000 TWh of electricity annually, more than twice the roughly 4,500 TWh produced by the U.S. Its power-generation capacity is also expanding nearly 11 times faster.