Genesis Mission, Scaling Limits, & TPUs

TLDR;

Welcome back! This week, we look at the US’s state-backed AI acceleration push, Google’s escalating TPU challenge, Ilya Sutskever’s argument that scaling is hitting its limits, the rapid shift from chat-based models to interactive AI products, including Google’s amazing Dynamic View feature, and more.

Enjoy.

POLITICS

US’s AI Manhattan Project?

The US Department of Energy has launched the Genesis Mission, a new initiative to double the pace of scientific discovery over the next decade by aggressively leveraging AI and advanced computing.

President Trump signed the executive order creating the program recently, building directly on his earlier AI Action Plan. The mission will unite all 17 DOE National Laboratories, the department’s supercomputers, and its quantum computing systems into a single coordinated platform.

More than 40,000 scientists, engineers, and private-sector partners will work together to build what the department calls “the world’s most powerful scientific instrument.”

Newly appointed Secretary for Science and Innovation Dario Gil, a former Spanish IBM executive confirmed by the Senate in October 2025, has been named to lead the effort.

The program will focus on accelerating breakthroughs in energy production, fundamental scientific discovery, nuclear advancement, defense materials, and national security. Officials describe the undertaking as the modern equivalent of the Manhattan Project, but with AI replacing the atomic bomb as the transformative technology. The Genesis Mission is already moving forward with full funding and operational authority granted by the executive order.

Out of the several invesments we are going to see, it seems Amazon is coming first, with an announced $50 billion on infrastructure for the US Government (at NVIDIA equivalents, that should be around 1 GW, but considering Amazon chips are cheaper and will be offered at great rates to the Government, the number is indeed larger, at ~1.3GW).

TheWhiteBox’s takeaway:

Right move… probably. We can, of course, take the cynical view and assume this is just laying the foundation of what seems to be the US Government's official intervention in the AI sector.

Here, the US acts as a “guaranteed customer” to Big Tech, which offers its infrastructure at very competitive prices, while also allowing the US Government to get in bed with them without raising too many flags.

You can agree or disagree all you want, but this was something that was going to happen, regardless of whether you believe the AI race deserves the same treatment as nuclear bombs.

The reason is that AI investment is already the only thing preventing the current administration from entering a recession a year from the midterms, so you can bet the US Government isn’t going to let AI go down, not for 2026 at least.

Will that liquidity be enough to sustain valuations over the next year? I don’t know, but you can bet this is not particularly bearish for AI stocks, if you get what I mean.

THEWHITEBOX

NVIDIA and Google… What to Make of it?

A new research note from Bernstein’s Stacey Rasgon outlines the main takeaways from the recent NVIDIA crash amid fears…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.