Andrew West
191 posts


Goldman just put a number on the AI buildout: $7.6T from 2026-2031.
They explicitly flag optics as the next “buy out the store” chokepoint after memory.
The numbers behind the thesis:
+Compute alone is $5.1T of the $7.6T total
+Annual AI capex grows from $765B in 2026 to $1.6T in 2031
+Data center cost per MW jumped from $10M (cloud era) to $15-20M (AI era)
+A single $50K accelerator depreciates $10K/yr but goes economically obsolete faster than the schedule
+NVIDIA’s GB300 NVL72 packs 72 processors per rack, linked by hundreds of thousands of km of cabling
In the repot they state: similar episodes of intense, short-term pricing pressure are likely to recur across other critical components such as interconnect, optics, storage, and packaging.”
Meaning the same dynamic that just sent memory parabolic is queued up across the entire physical layer of AI infrastructure.
The “AI factory of the future” data center will pack 576 GPUs per rack at 500+ kW, requiring liquid-only cooling and millions of GPUs deployed at the >1 GW scale.
None of that scales without optics.
Memory just ran (still going). Photonics is just warming up.
Full disclosure:
Goldman is confirming what’s been playing since the late last year. A lot is already priced in.
The alpha is front-running institutions, not reading their reports.
But $7.6T over 6 years doesn’t get fully priced in 6 months. Plenty of runaway left.

English

@OpenAINewsroom @ChatGPTapp @grok @grok I paid up for my blue check now pls answer!?! Is this OAI real account? Confirmed elsewhere??
English

@OpenAINewsroom @ChatGPTapp @grok is this account really managed by OpenAI? have they confirmed this elsewhere?
English

In January 2025, we committed to generating 10GW of compute and have already identified over 8GW of that. Now, we're planning for 30GW of compute by 2030. A milestone that scales with the rapidly accelerating demand for intelligent systems.
Image generated by @ChatGPTapp Images 2.0 😉 #YouCanJustBuildThings

English

At a conference last week, the heads of OpenAI infra said all type of their workloads (incl. training + inference combined) could in theory interruptible - just that they prefer not to.
Tech companies are definitely working on this behind the scenes on all these constraints. Excited to see developments on interruptibility, temporal or geographical flexibility, distributed training, and hardware/software/BESS at GPU and rack level!
Duncan S. Campbell@duncancampbell
Interesting also that he says AI training centers can handle outages and are sometimes skipping installing backup generators.
English

@ShanuMathew93 $CEG will own the site if its CPN acquisition goes through. Does the wording here imply that ECP will build/own the data center even if it no longer owns the power gen / site? Seems odd.
English

Usually pretty good at catching all the data center news headlines but scanned the past few weeks and think I missed two of these
1) Energy Capital Partners (ECP) plans to develop a $5 billion data center through Calpine Corporation at the 828MW York 2 Energy Center in Peach Bottom Township, York County. The project is expected to create 2,500 construction jobs and is part of a $90 billion investment in Pennsylvania's digital infrastructure.
2) Skybox Datacenters is building a $125 million data center as the second building in its PowerCampus Austin project in Hutto. Construction begins October 1, 2025, with completion by June 2026, part of a 600MW campus offering up to 4 million sq ft.


English

@NicNack228 The part about not wanting to open the detergent really speaks to me
English

@salemisms Gotta do it. It’s the best. Then we can do yoga together namaste 🙏
English

@StatisticUrban Could you recut as a ratio of median house price : median HH income ratio? @StatisticUrban
English

A little bit more granular - linear distribution of house prices by county. There's a decent case to be made that the US housing crisis is really a housing crisis of DC, Boston, NYC, Colorado, Florida, and the west coast.
Of course, that's also the areas with high demand...

Hunter📈🌈📊@StatisticUrban
The scale of the housing crisis in Hawaii, California, Massachusetts, and New York becomes quite apparent if you do a linear distribution of house prices by sq ft. They leave all other states in the dust.
English

@FoolAllTheTime If you think COTY is interesting just wait until you can buy Revlon post reorg equity
English



