Michael Kratsios, Maria Bartiromo: Inside America's AI Strategy: Infrastructure, Regulation, and Global Competition

The conversation opens with an assessment of where the United States stands in its effort to be the leading AI country. The answer offered is confident: the U.S. is "doing great." The framing goes back to a major AI policy speech President Trump gave in July of the previous year, in which he declared that the United States was in an AI race and had to win it. The panelist compares that speech to President Kennedy's declaration of a space race.

The case for optimism rests on the pace of American products. The panelist says American companies have "only innovated more" since the speech, and that American models, chips and data centers keep improving. The same answer concedes that the U.S. faces "very, you know, competent and formidable competitors," and that China has "a lot of very smart people working in this area."

The interviewer then raises a recurring doubt: is too much being spent on data centers, and will the investment pay off? The response is that the build-out reflects real demand. The panelist anticipates the comparison with the late-1990s fiber build-out and the dot-com crash that followed, and draws a distinction. Back then, much of the fiber that was laid went unused, a problem known as "dark fiber."

There's no such thing as a dark GPU right now. Every GPU that's being put in a data center is getting used.

According to the panelist, those GPUs generate tokens for AI chatbots and coding assistants. Recent coding releases are described, secondhand from software developers, as "mind-blowing" and "completely revolutionizing their industry." Rising token demand drives demand for more data centers. The panelist also puts a number on the macroeconomic effect: last year the infrastructure build-out added "about 2% to the GDP growth rate," helping produce what is described as a four to five percent growth rate, and something similar is expected this year. No source is given for these figures.

A second participant summarizes the administration's AI plan as resting on three pillars, and says the federal government has made "pretty good progress on all three":

Out-innovating U.S. competitors. Driving the infrastructure build-out needed to support the AI revolution. Sharing, or exporting, American technology with the world.

On innovation, the key point is regulation. The view expressed is that innovation needs a regulatory environment that lets the technology be developed and commercialized in the United States. The U.S. is said to have done well compared with the rest of the world but could still improve. The central concern is the prospect of 50 different state rulebooks for AI, an issue the president raised in his July speech. The speaker argues that a patchwork hurts newcomers more than incumbents.

the patchwork is actually most detrimental to early stage young companies and entrepreneurs

The reasoning is that a startup building on a frontier model faces heavy friction if it must navigate rules across 50 states, while "the big guys are the ones that can succeed in that environment the best." The administration is working on a legislative proposal for what is called a "sensible national framework."

The interviewer notes that some states have pushed back and want to control their own AI destiny. In reply, the panel says a December executive order directing work on the proposal listed areas states should keep regulating on their own, including child safety and the permitting of data centers and build-outs. The panel does not give the full list and says these details are still being worked through.

The criticism of state activity is blunt. One panelist says "the states are going hogwild right now with regulation," citing over 1,200 bills in state legislatures and calling it "very much a knee-jerk reaction." The argument is that every hypothetical concern now attracts several bills before anyone knows how the technology will play out. Because the technology is new and the environment dynamic, the panelist says it would be better to study how AI is actually used and which risks actually appear before regulating heavily.