The U.S.-China Economic and Security Review Commission published a paper in March 2026 titled "Two Loops." It analyzes how China's open-source AI strategy creates a faster iteration cycle than America's proprietary approach. The two loops are digital (open-source distribution drives adoption, adoption generates training signal, signal improves the next model) and physical (manufacturing dominance generates real-world data that feeds embodied AI). The intersection of these loops, the Commission wrote, gives China's strategy its compounding force.
The title is a reference the authors may not have intended. John Boyd spent his career arguing that the faster loop wins.
Boyd was a fighter pilot at Nellis Air Force Base in the late 1950s. He earned the nickname "Forty-Second Boyd" because he had a standing bet: $40 to any pilot who could defeat him in simulated air combat within forty seconds, starting from a position of disadvantage. He usually won in twenty. Students, instructors, Marine and Navy pilots, officers from a dozen countries attending the Fighter Weapons School under the Mutual Defense Assistance Pact. No one ever collected.
His contribution to the Air Force wasn't the flying. It was a theory of flying. In the 1960s he developed energy-maneuverability theory, a mathematical framework for comparing fighter aircraft based on their energy state at any point in a maneuver. The theory reshaped how the Air Force designed the F-15 and F-16. Then he moved on.
What he moved on to was a set of briefings. "Patterns of Conflict" ran up to six hours. He never published it. He presented it in Pentagon conference rooms, at war colleges, to anyone who would sit through it. The ideas spread by photocopy, then by word of mouth, then through the people who sat in the room and carried the framework into their own work.
The OODA loop was the core. Observe, orient, decide, act. But the loop itself isn't the insight, and most people who cite it miss what Boyd actually emphasized. The critical step is orient. Orientation is the lens through which you interpret what you observe. It's shaped by genetics, experience, cultural traditions, previous learning. Two pilots observe the same dogfight. The one with better orientation sees different possibilities. Orientation is why two people can look at the same data and see opposite conclusions. It is the step most people skip when they draw the OODA loop on a whiteboard.
Boyd's argument was about tempo, not speed. The fighter that cycles through the loop faster forces the slower fighter into a reactive posture. The slower pilot responds to a situation that has already changed by the time he acts on it. Each cycle widens the gap. Eventually the slower pilot is orienting to a world that no longer exists. He doesn't lose because he's less skilled. He loses because his model of reality is lagging his opponent's.
Boyd applied this first to air combat, then to ground warfare, then to grand strategy. The framework anticipated the logic of Desert Storm years before it happened. When the coalition used speed and dislocation rather than mass and attrition, the Iraqi army collapsed in a hundred hours. Observers called it surprising. Boyd's acolytes in the Marine Corps did not.
Boyd retired as a Colonel. He never held a general officer's command. He published one short paper, "Destruction and Creation." His most important work survived as briefing slides and the notes of people who attended. Robert Coram's biography, published five years after Boyd's death in 1997, introduced him to readers who had never been in the room. The man who taught that tempo beats mass left a body of work that can barely be cited.
The USCC paper maps his logic onto AI competition. Chinese labs release models openly, absorb improvements from the global developer community, and iterate. DeepSeek V4 Pro matches frontier proprietary models on most agentic benchmarks at roughly 34 times lower API cost per output token. Roughly four out of five U.S. AI startups that use open-source models choose Chinese ones. The open loop is faster.
The proprietary labs operate a different cycle. Closed development, internal iteration, gated release. The models may score higher on any given benchmark. But scoring higher on a benchmark is not the same as cycling faster. A better model released quarterly loses to a good-enough model that improves weekly, for the same reason a better pilot who orients slowly loses to a faster one who orients quickly.
Boyd would recognize the shape. The question isn't which model tops a leaderboard. It's which system reorients faster when the ground shifts.
His final briefing slide asked a question he put to every young officer who came to him for career advice: "To be or to do?" He meant it as a choice between pursuing rank and pursuing influence. He chose influence. He died without the star that would have made his obituary front-page news.
The question works for AI strategy too. Build the strongest model, or build the fastest loop.