Eighty Years to Build, One Thursday to Lose
America invented the computer, the internet, and the AI that runs on both. Beijing just caught us—and is about to give the frontier away for free.
by Lawrence Winnerman, Blue Amp Media COO
Once you’ve finished this, read the companion piece, The Layer War.
On Thursday, July 16th, a company most Americans have never heard of released an artificial intelligence model that caught the American frontier lab models, and it released it from Beijing.
The company is called Moonshot AI. The model is Kimi K3: 2.8 trillion parameters, the largest model ever slated for open release, and within hours of its debut it stood in first place on Arena’s Frontend Code leaderboard, ahead of Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol, the two most capable systems the American labs have ever shipped—first, in fact, in six of the seven coding categories. On Arena’s broader text rankings it finished ahead of Claude Opus 4.8, Anthropic’s flagship until five weeks ago, while costing 40 percent less to run.
And on Monday, July 27th, Moonshot says, it will give the whole thing away: the weights, the architecture, everything, free to download and free to run in any data center on earth. The weights are the model itself, the billions of numbers in which everything it learned is stored; whoever holds them holds the machine.
Moonshot—and, by extension, China, just beat us at AI and is giving the model away. For FREE.
Axios said it plainly Friday morning: “China just erased America’s AI lead.”
Eighty years to build, and all it took was one random Thursday to lose.
The computer is an American invention.
The theory owes a debt to a British mathematician, Turing, who his own government hounded to death, but the machines—the actual, humming, room-filling, world-changing machines—were built here in the United States.
In February 1946, in a room at the University of Pennsylvania, engineers threw the switches on ENIAC, thirty tons of vacuum tubes and hand-soldered wire that could do in thirty seconds what a human being with a desk calculator needed twenty hours to finish. The transistor came out of Bell Labs in New Jersey the following year; the integrated circuit out of Texas and California a decade after that; the microprocessor out of Santa Clara in 1971.
The supercomputer was substantially the work of one stubborn man in Chippewa Falls, Wisconsin—Seymour Cray, who treated speed as a calling. The internet began as ARPANET, a Pentagon research project that spoke its first halting word between UCLA and the Stanford Research Institute in the autumn of 1969. And the transformer architecture that makes every modern AI model possible—Claude, ChatGPT, Gemini, and yes, Kimi K3—was published in 2017 by eight researchers at Google, in a paper the company let out the door for free.
Every layer of this revolution was laid down on American ground, much of it with public money, much of it by people who came here from somewhere else because here was where you came to build. The record reads like myth and is footnoted like history. Computing is the American inheritance the way navigation belonged to Portugal and steel belonged to Sheffield, except bigger, because computing became the substrate everything else now runs on—including steel manufacturing and navigation.
I am not a neutral witness to any of this. I spent twenty-five years inside the American tech industry, managing programs at the largest scale it had to offer, and I believed—actually believed, the way you believe things at two in the morning in a war room when the release finally goes green—that we were building the future for everyone.
Then the industry decided that people my age were a cost center, the first wave of this technology arrived to absorb our work, and I got to experience the American way of implementing AI from the receiving end.
I know what these companies are from the inside, which is why you should believe me when I tell you: this was the home game, the one field we were supposed to lose last, or never at all.
In 2023, the men running America’s AI industry told us China was eighteen months behind, and eighteen months, at frontier speed, was forever.
By last year the number was twelve months, maybe six. This April the federal government’s own AI testing center called it roughly eight months, and Stanford’s AI Index, reading the leaderboards this spring, found the gap between the top American and Chinese models had shrunk to 2.7 percent—no longer a lead so much as a rounding error with good publicists. Yesterday the number arrived at approximately zero, carrying a price tag 40 percent lower than ours and a shipping date for the free copy.
Eighteen, twelve, eight, two-point-seven, zero: a countdown every one of these executives could read the whole time, while they assured Congress and their investors and the rest of us that the moat was real.
We had mistaken being early for being ahead.
The rest of this essay names names: who sold the American century, why they sold it, and what taking it back actually looks like. Blue Amp Media takes no money from the men this piece indicts—we're funded entirely by readers. Join the people paying for independent progressive media, and read the fight plan.
