THE POST
What it’s like trying to rent AI/SI compute https://t.co/SBdEXK31zc
ENGAGEMENT
- likes10,029
- reposts1,138
- replies1,524
- quotes272
- bookmarks1,155
- views2,562,196
AS RETURNED BY THE X API · FETCHED 2026-10-19 05:20 UTC
WHAT IT SAYS
The caption says the attached clip shows what it is like to try to rent AI or SI (superintelligence) compute. The 79-second video is a comedy sketch set in an old wood-panelled shop: a young customer talks to a shopkeeper in a bow tie and apron, a musician plays a stringed instrument in the corner, and the shelves behind the counter hold server racks instead of goods. I viewed still frames only, without audio, so I can't say what is said; the post's point, carried by the caption, is that AI compute is hard to get.
CONTEXT
The scarcity the post jokes about shows up in the power numbers. Morgan Stanley's report "What Does the Power Shortfall Mean for Semis?", as covered by The Economic Times, estimates a 32 GW US data-centre power shortfall through 2028, about 34% of projected needs, possibly widening to 68 GW cumulative by 2029, and quotes the bank saying there are more semiconductors being built than places to put them. It also expects big, creditworthy players to take most new capacity, which could squeeze smaller buyers of compute.
A live example from the same weekend: TNW, citing the Telegraph, reports that Microsoft pulled out of anchoring Nscale's planned data centre at Loughton, Essex, which has been told not to expect enough grid power until the 2030s. Nscale calls it a London-wide grid issue and says it is looking at other power sources and sites.
The other side of the ledger is new generation built for compute. Georgia's Public Service Commission approved a contract for Georgia Power to build 3.2 GW of new capacity for OpenAI's planned Effingham County campus, with OpenAI paying for the infrastructure, offering up to 1 GW of flexible demand response, and deliveries expected between 2028 and 2032.
WHAT IT MEANS FOR K
Neutral. A joke doesn't add or remove a watt. But the shortage it points at is a real limit on the climb: chips are arriving faster than powered buildings to run them, so the pace at which compute turns into harnessed power is set by generation and grid connections. That shortage is also the price signal pulling deals like Georgia's 3.2 GW into existence.
CLAIMS
- Renting AI compute is hard right now (implied by the caption and the sketch).
EVIDENCE
- Morgan Stanley, as reported, sees a 32 GW US data-centre power shortfall through 2028, about 34% of projected needs.
- Microsoft reportedly left Nscale's Loughton project after the site was told grid power would not arrive until the 2030s.
- Georgia regulators approved 3.2 GW of new generation for one OpenAI campus, with power expected from 2028.
REASONING
- The evidence supports the post's premise that compute is scarce and that power is a main cause.
- The post itself is commentary and changes nothing about how much power is generated.
- Under this site's rules, a post with no effect on harnessed power is neutral, even when the problem it points at matters a lot for K.
SEE ON THIS SITE
TAGS
- COMPUTE
- DATA CENTRES
- POWER DEMAND
- GRID