US AI Data Centre Timelines Are Looking Shaky, and SEA Should Be Watching
The AI gold rush is still going hard, but the actual buildings behind that hype might be moving slower than expected.
According to a Financial Times report citing geospatial analytics firm SynMax, at least 40% of AI data centres planned for completion in 2026 could be delayed by more than three months. The problem is not some dramatic collapse. It is the usual real-world headache combo: regulatory friction, supply chain issues, labour shortages, and not enough power infrastructure ready in time.
SynMax is not just guessing here. The company uses satellite imagery and AI to track construction progress, then checks that against public statements, permit records, regulatory documents, industry intel, and interviews on the ground. So instead of trusting glossy corporate updates, it looks at what is actually happening at the sites.
One of the clearest examples is a huge Oracle-built campus in Shackelford County, Texas, which is meant to support OpenAI. The project spans 1,200 acres, includes 10 buildings, and is expected to deliver 1.4 gigawatts of capacity. Officially, the site is still targeting the second half of 2026. But satellite images from early April show only six plots cleared, and just one of them appears to be seeing meaningful development. SynMax's view is that maybe one building could be ready by the end of this year, while a more realistic broader rollout slips into 2027.
Another OpenAI-linked site in Milam County, Texas, is also reportedly progressing slowly. That one is expected to reach 1.2GW, but imagery seen by SynMax shows only one building under construction so far.
The companies involved are pushing back hard. OpenAI said its data centre expansion is on schedule and will accelerate, while Oracle said each site it is developing for OpenAI is moving forward on time. SB Energy also said the Milam County facility remains on schedule and could still become one of the fastest data centres of its kind ever delivered.
Still, people closer to the ground are telling a less confident story. Construction executives say specialist workers, especially electricians and pipe fitters, are in short supply. That tracks with reports stretching back to late 2025. And these OpenAI-linked builds are apparently not the only ones under pressure, with another recent report claiming that half of planned US data centre projects have either been delayed or cancelled.
Power is another massive issue. Building a shell for AI GPUs is one thing. Feeding those systems enough electricity is another. Even if the hyperscalers are paying for grid upgrades, utilities still need time to order equipment, deploy it, and finish the supporting infrastructure. Some companies are now looking at on-site turbines and generators, but those come with EPA permit requirements. On top of that, jet engine supply chains are also tight, with some 2025 orders reportedly not landing until 2028 to 2030.
Why should readers in Malaysia and SEA care? Simple: the AI race is global, and many of the tools we use, from cloud AI services to game dev workflows, localisation systems, moderation tools, and esports broadcast tech, depend on backend capacity scaling properly. If the biggest US projects start slipping, rollout timelines for AI-heavy services could get messier too. For SEA, this is also a reminder that building AI infrastructure is not just about announcing billion-dollar plans. You still need land, labour, permits, parts, and serious power.
So no, this does not mean those data centres are dead. But it does mean the AI boom is running into the boring physical limits of the real world, and investors expecting fast returns may need to chill a bit.
Source: Tom's Hardware
