Last week, the White House announced Project Genesis, a new moonshot initiative to coordinate leading AI companies to accelerate and control the development of AGI. Many commentators have likened this to a new “Manhattan Project,” but a student of history would immediately see the glaring differences between Genesis and Manhattan: (1) Genesis was publicly announced whereas Manhattan was secret, (2) Genesis is trying to coordinate diffuse talent with loose alignment whereas Manhattan concentrated top researchers physically in one location, and (3) Genesis has no explicit funding whereas Manhattan consumed 0.3% of GDP—equivalent to $84 billion per year for the US, or $60 billion per year for China.

The only part of Genesis that looks vaguely like the Manhattan Project is the fact that it’s being coordinated by the Department of Energy. This makes strategic sense: energy is the binding constraint on US AI development. In emergency situations—say, if US intelligence obtained credible information that China had achieved AGI—Genesis with DoE support could theoretically ration all non-AI energy consumption to accelerate development. But this is a catch-up scenario, not a path to leadership. It is hard to see how Genesis as currently described can dramatically accelerate AGI development given its publicity, diffusivity, and lack of committed resources.

In contrast, I estimate there’s a 60-80% chance that the equivalent of an AGI Manhattan Project is currently happening in China right now.

The Project 596 Precedent

China, as a centralized command economy and political system, is extraordinarily good at 5-year moonshot projects when convinced to do so. The best evidence that I may be underestimating with my 60-80% estimate is that China actually did something very similar in the early 1960s for developing its nuclear program, under “Project 596.”

The project was named after June 1959—the month Khrushchev decided to stop helping China with nuclear development. The Soviets withdrew over 1,400 advisers and technicians, taking their blueprints with them. The Central Committee’s response: “Do it yourself, start from scratch, and prepare to build the atomic bomb in eight years.” They deliberately named the project after the date of this “national disgrace” to fuel motivation.

In an era when millions were starving in the Great Leap Forward famine, China diverted a full 1.0% of its GDP to nuclear development. Over 10,000 scientists and technicians were mobilized, many of whom “disappeared” from public life—changing their names, cutting all foreign contact, becoming invisible for more than a decade. The result: China’s first atomic bomb test in October 1964, and its first thermonuclear bomb (3.3 megatons) just 32 months later. For comparison, the United States took 86 months to go from fission to fusion. China achieved this while isolated, impoverished, and politically chaotic.

It is not only feasible for China to launch secret national priority projects—it is a tradition.

The “Project 1023” Hypothesis

If China is running an AGI Manhattan Project today, what might it look like? The October 2023 US chip export controls—when Washington ordered Nvidia to immediately halt exports of the H800, closing the workaround China had relied on after the initial 2022 restrictions—provide a natural “596 moment.” That’s when Chinese leadership realized the US would keep tightening the noose on any compliant chip. Call it “Project 1023.”

China’s advantages in such a program are substantial. It can actually enforce secrecy—the US leaks everything, including the Genesis draft before it was signed. It can concentrate resources without shareholder demands or antitrust concerns. It can “disappear” researchers from public life without investigative journalists asking questions. And critically, the efficiency gains demonstrated by DeepSeek (achieving near-frontier performance at reportedly 10-20% of Western training costs) suggest that Chinese labs may have algorithmic advantages that reduce their compute requirements.

The timeline fits. Two years have elapsed since October 2023—enough time for a serious program to be well underway. If planning began in late 2023 or early 2024, China could plausibly have gathered 70% of its top 500 AI researchers into a classified program by now.

Where Would It Be?

A secret AGI facility requires massive power (gigawatts, not megawatts), cooling infrastructure, network connectivity, talent willing to relocate, and a plausible cover story. Following the energy tells us where to look.

Guizhou Province is my top candidate. China has already built massive data center complexes there—Apple’s iCloud China data is hosted in Guizhou. The province has abundant hydroelectric power from the Wujiang River system, cool mountain climate for reduced cooling costs, existing fiber optic backbone, and is already normalized as a data center hub. New construction wouldn’t raise flags. A secret facility could hide within or adjacent to the existing Gui’an New Area data center cluster. Satellite imagery would show “more data centers in the place that already has data centers.”

Sichuan Province follows similar logic—massive hydroelectric capacity from Three Gorges tributaries and Jinsha River projects. Chengdu is a legitimate tech hub with universities and existing AI talent. A facility in the mountains west of Chengdu could draw power from hydro while maintaining talent access.

Xinjiang is counterintuitively less likely despite its remoteness. It’s already under heavy US surveillance for Uyghur human rights monitoring, and the region has genuine infrastructure limitations. However, the massive new solar installations in western Xinjiang could provide cover for anomalous energy consumption if China were willing to accept the surveillance risk.

Unlike some Western analysts who assume China would use a distributed network architecture, a true Manhattan Project equivalent would almost certainly be physically centralized. The reason is cryptographic risk: with Google’s recent Willow quantum computing advances and the possibility of classified quantum projects further ahead, any distributed system relying on encrypted communication between sites faces the risk of future decryption. A centralized facility with air-gapped systems eliminates this vulnerability entirely. The original Project 596 was centralized at Lop Nur and the Northwest Nuclear Weapons Research and Design Academy for the same fundamental reason—compartmentalization works best when the compartments are rooms in one building, not nodes on a network.

The Detection Problem

US intelligence presumably tracks Chinese AI capabilities, but the uncertainty bands must be enormous. Unlike nuclear weapons development, there’s no equivalent to radioactive signatures or centrifuge procurement. Talent can be “disappeared” gradually (researchers just stop publishing, attributed to industry jobs). Compute can be distributed across multiple facilities. Progress isn’t marked by physical tests—there’s no equivalent to Trinity that seismographs can detect.

US intelligence knew China was pursuing nuclear weapons in the early 1960s but consistently underestimated the timeline. The hydrogen bomb came shockingly fast. If China is running a “Project 1023,” would US intelligence be any better at estimating their AGI timeline than they were at estimating 596?

The 596 Mindset

Does the current Chinese leadership have the same capacity for sustained commitment that characterized the 596 era? Recent history suggests yes. Chinese leadership has demonstrated willingness to accept significant economic costs—COVID lockdowns, tech sector restructuring, real estate market corrections—for strategic goals. The CCP’s legitimacy narrative increasingly emphasizes “national rejuvenation” and overcoming historical challenges. The chip export controls fit naturally into that narrative frame.

Project 596 succeeded through total state commitment (1% GDP during famine), focused prioritization (protecting key scientists even during the Cultural Revolution), acceptance of significant costs, and channeling national determination toward a clear goal. These institutional capabilities remain intact.

Genesis signals intent; Project 1023—if it exists—would represent execution.

---

By David Zhang and Claude Opus 4.5

November 28, 2025

© 2025 David Yu Zhang. This article is licensed under Creative Commons CC-BY 4.0. Feel free to share and adapt with attribution.