Frontier AI infrastructure (data centers, memory supply) will not reach adequate scale until late 2028 or early 2029.

The AI hardware bottleneck is now memory and bandwidth rather than compute.

“We are not in a computebound world. We are in a memory and network and communication or bandwidthbound world.”

TSMC's reduced growth rate in 2023-2025 will worsen the compute shortage over the next few years because fab lead times are longer than data center lead times.

“TSMC decreased their rate of growth in 2023 and 2024 and 2025. So like we're our shortage of compute is going to get worse in the next few years because a fab the lead time is even greater than a data center.”

Amp Infrastructure expects to need approximately 6 gigawatts of spike compute capacity over the next 4 years to support frontier AI teams.

“I think the steady state would be that we have a base load pool of 1.3 gigawatts at all times of base load capacity. For spike capacity, right now my estimate is we need roughly 6 gigawatts over the next 4 years for all our teams.”

As coding models accelerate implementation speed, compute will again become the primary bottleneck for AI research iteration.

“Now coding models are much more efficient and can help us implement stuff much faster. compute might become a bottleneck again because previously like if you want to train a new model say you want to generate new synthetic data and then or write a new algorithm it might take a few weeks and during that period of time you don't you might not have experiments to run but now you can build that thing within a few hours then you can immediately train a model now you have to have enough compute to try all of the ideas. So compute might be the bottleneck of iterating speed again.”

Data center capacity at scale will not be available until late 2028 or early 2029.

“We're massively supply constrained. You can't get data center capacity at scale >> until late 28, early 29 right now. And that's just a fact.”

Compute scaling will slow down in the late 2020s due to fab capacity constraints.

“Compute scaling will start to slow down probably in the late 20s because just like at some point you use up basically almost there.”

Five more years of AI progress would enable an AI to take over a fab and produce more chips.

“I'm going to take over a fab and produce more chips. I'm going to go into Congress and try to convince them to pass some bill.”

The AI data center power shortage will begin to alleviate in 2027-2028.

“I think the watts shortage will probably begin to alleviate 27 28.”

Nvidia's Blackwell GPU generation provides scale-up memory on the order of 10-20 terabytes, which is sufficient to hold a ~5 trillion parameter model plus KV cache.

“With Blackwell finally, which was deployed in… Maybe last year. You finally have a scale-up on the order of 10-20 terabytes, which is enough for a 5T model plus KV cache.”

By 2027, AI progress will be primarily proportional to compute, and human researcher talent will no longer matter significantly.

“One of the key aspects of AI 2027 the tabletop is that mostly your progress is proportional to your computer And your current place like they're like you move up the timeline at a different at a at a rate proportional to what percentage of the world's compute you have because your researchers no longer matter very much, right?”