The next phase in the AI race is going to look different: It will be defined more by physical construction than by scientific discovery. (View Highlight)
Up until now, you could fit your training cluster into an existing data center via colocation or retrofit. If you needed to increase cluster size from 15k GPUs to 25k GPUs, you found a way to plug-in more GPUs. This is changing: The “Bitter Lesson”—which most market participants in AI have internalized—says that model size is the number one driver of performance. As a result, the next generation of models are aiming for a 10x increase in model scale to 300k GPUs. To house one of these models, you need to build an entire new data center. (View Highlight)
This changes AI in two fundamental ways: First, it changes the lead time between models. If before you could train your model in 6 to 12 months, now you need to add 18 to 24 months of construction time before you can actually start training. Second, it changes the source of maximum competitive advantage. In the new era, construction efficiency may matter more than research breakthroughs. (View Highlight)
This sea change in how AI works was a major theme of big tech earnings last week. Annualized CapEx for big tech increased from 138Bto229B year-over-year. This incremental $91B in run-rate spending is a good proxy for new AI data center construction—an enormous investment. (View Highlight)
Today’s CapEx will likely yield fruit somewhere between late 2025 and early 2026, at which point we’ll find out if these larger models are intelligent enough to unlock new revenue streams and generate a return on investment. (View Highlight)
So what exactly is going to happen over the next 1 to 2 years, and how does one “win” in this new phase of AI? (View Highlight)
Today, five companies have arrived at the starting line in this new race toward data center scale-up: Microsoft/OpenAI, Amazon/Anthropic, Google, Meta and xAI. Each has a model that has held up against serious benchmarks, and the necessary capital to proceed. (View Highlight)
Meta and xAI are consumer companies, and they will both vertically integrate, hoping to benefit from each having a single founder decision maker who can streamline and tightly couple model building efforts with data center design and construction. Both companies will seek to launch killer consumer applications on the back of more intelligent models. (View Highlight)
Microsoft and Amazon have grizzled data center teams and deep pockets, and they’ve leveraged these assets to forge partnerships with the top tier research labs. They hope to monetize through 1) Selling training to other companies, and 2) Selling model inference. They will need to manage resource allocation between their frontier models (GPT 5 and Claude 4) and other data centers being built for Enterprise customer use. (View Highlight)
Google has both a consumer business and a cloud business, and also has its own in-house research team. On Friday, the company announced it was bringing Noam Shazeer back into the fold. Google also has vertically integrated all the way down to the chip layer with TPUs. These factors should provide long-term structural advantages. (View Highlight)
With CapEx plans now firmly in place and the competitive landscape set, the new AI era begins. In this new phase of AI, steel, servers and power will replace models, compute and data as the “must-wins” for anyone hoping to pull ahead. (View Highlight)
