George Gilder published Gaming AI: Why AI Can’t Think But Can Transform Jobs in 2020. The book arrived just before the large language model explosion, and it argued the opposite of what the entire tech industry was about to spend a trillion dollars believing.
The thesis was simple. Machines process patterns, and humans create. Those activities are not equivalent, and confusing them is the central error of the AI era.
Gilder spent the next five years on the public circuit pushing that argument, and the public circuit kept growing. By 2026, the Gaming AI thesis was the intellectual framing for his wafer-scale computing thesis, his Convergence X investment campaigns, and a guest essay in Porter Stansberry’s Daily Journal. The prophet of the free economy and author of Telecosm — a man who had spent three decades arguing that networks beat processors — was now arguing that minds beat machines. The framework was the same. The application was new.
The Core Argument
Gaming AI has three load-bearing claims.
First, AI cannot think. Gilder distinguishes between processing and thinking. A machine can sort a billion images and identify the ones that contain cats. That is pattern recognition. It can generate a plausible sentence by predicting the next token based on a corpus of billions of prior sentences. That is statistical simulation. Neither activity is thinking in the sense that a physicist thinking through a new theory is thinking. The physicist creates; the machine rearranges.
Second, the human brain runs on roughly 12 watts of power, while a modern AI training run requires gigawatts. The gap is not a footnote. It is the central evidence for the first claim. If statistical pattern matching were all that thinking was, the energy budget would not diverge by six orders of magnitude. The 12-watt brain is doing something the gigawatt data center is not, and that something is what Gilder calls creativity.
Third, AI will still transform jobs. This is the part that gets lost in the summary. Gilder is not a doomer, and he does not argue that AI is useless. He argues that AI is a tool, like the plow or the loom, and tools displace specific kinds of labor while creating demand for new kinds. The jobs that disappear are the ones that consisted of pattern processing. The jobs that survive and grow are the ones that consist of genuine insight, imagination, and judgment.
Where the Book Came From
Gilder’s AI skepticism did not appear in 2020. It has roots in every book he has written.
Wealth and Poverty (1981) argued that wealth comes from entrepreneurship and creativity, not from the redistribution of existing resources. The supply-side economics that made Gilder Reagan’s most-quoted author was built on the premise that the human mind is the source of all economic value. You cannot redistribute what has not been created, and creation is what minds do.
Microcosm (1989) extended the argument to technology. The microchip was the abundance of the era, but the meaning of the microchip was that it let human creativity scale. The PC was a creativity amplifier, not a creativity replacement.
Knowledge and Power (2013) applied Claude Shannon’s information theory to economics. Information, in Shannon’s sense, is surprise. A message that tells you what you already knew contains zero information. Growth comes from unexpected discoveries, and unexpected discoveries come from minds that generate things no algorithm could have predicted from the prior distribution. A system that runs on surprise cannot be run by a machine that extrapolates from the past. The five-decade arc from Microcosm forward traces the method that keeps producing these calls.
Life After Google (2018) moved the argument to the architecture of the internet itself. The blockchain thesis argued that centralized data monopolies were architecturally exhausted — the same creativity-vs-pattern-processing distinction applied to networks rather than chips.
Gaming AI is the application of that framework to the specific question of whether machines can think. The answer was always going to be no, because Gilder’s entire intellectual project depends on the answer being no. If machines could think, the creativity thesis at the center of his economics would collapse. The book is a logical consequence of everything he had written for four decades.
What Time Did to the Thesis
The hard question is whether the book has aged well, and the honest answer is that it has aged into a more interesting argument than the one Gilder made.
The 12-watt-brain claim is the strongest part of the book and still stands. No one has explained why a 175-billion-parameter language model requires megawatts to train and the brain that designed it requires a tuna sandwich. The energy gap is real, and it is the kind of fact that survives whatever the benchmark-of-the-week says about model capabilities. If you believe the brain is doing something more than next-token prediction, the energy budget is your evidence.
The “AI cannot think” claim is where the book is most contested, and Gilder is honest about the contest. He defines thinking narrowly, as the generation of genuinely novel ideas. By that definition, no current AI thinks. By a broader definition that includes useful simulation, reasoning over novel contexts, and the production of outputs no human had produced before, the question gets harder. The position in 2026 is that the line between simulation and thinking has blurred in ways that did not exist in 2020, and that blurring is the actual subject of the debate the book started.
The job-transformation claim is where the book has held up best. Gilder said AI would displace pattern-processing work and create demand for judgment work. That is what happened. The clerical and coding jobs that consisted of moving structured information from one format to another were the first to compress. The roles that involve genuinely novel design, novel strategy, and human relationship are the ones the labor market is still paying premiums for. The book did not predict the specific shape of the displacement, but the pattern it described is the one the data has followed.
The 2026 Resurfacing
Gilder spent 2026 pushing the Gaming AI argument across three high-profile venues, and the venues matter because they show the thesis has become the connective tissue of his current work.
The Porter & Co. Daily Journal essay (June 2026, “Illusions of Pure Reason”) was the sharpest statement. Gilder took the whole issue. The argument was the same one from 2020 with a new frame: human minds are the only real quantum computers, meaning they are the only systems that generate genuinely unpredictable outputs from deterministic inputs. The framing is borrowed from quantum information theory, and the physics is arguable, but the rhetorical move is clear. If minds are quantum systems, then no classical computer can replicate them, and the AI-vs-mind question is settled at the level of physics rather than benchmarks.
FreedomFest 2026 (July, Las Vegas) gave Gilder three sessions. “Reality Check for AI” on July 8 was the Gaming AI thesis in panel form. “The End of Chips” on July 9 was the wafer-scale thesis, the technological layer of the Gaming AI argument: if minds beat machines, then the machine side needs to get dramatically more efficient, and wafer-scale is the path to that efficiency. “AI: Artificial, Advanced or Alarming?” on July 10 closed the loop with a panel debate.
The America Out Loud PULSE podcast (June 20, with Dr. Randall Bock) was the popular-audience version. The 12-watt brain, the gigawatt data center, the creativity thesis, and the connection to Gilder’s investment newsletters. The audience for the appearance is the audience for the Moonshots subscription, which is the audience for the Ambient AI and Convergence X campaigns. The intellectual argument and the marketing engine run on the same fuel.
Where the Thesis Meets Its Edge
The strongest line of argument against Gaming AI does not contest the brain’s energy budget. It contests what the energy budget proves. The brain runs on 12 watts because it was built by four billion years of evolutionary pressure on energy efficiency. Silicon was not. If the energy gap is an artifact of how computers are currently built rather than a fundamental limit on computation, then the 12-watt argument establishes less than the thesis needs. A future architecture that closes the energy gap by three orders of magnitude would not, by itself, prove that the resulting system thinks. But it would remove the strongest piece of physical evidence Gilder has for saying it does not.
The same edge runs along the creativity thesis. “Genuinely novel” is doing work in the definition. If an AI produces an output that no human has produced before, and the output is useful, and no human could have produced it without the AI, the question of whether the output represents thinking or simulation starts to sound like a philosophical distinction rather than an empirical one. Gilder’s response is that the distinction is the whole point, and that losing it would mean losing the basis for treating human creativity as the source of economic value.
The debate the book started has not closed. Gilder’s 2020 argument was a bet that the line between simulation and thinking would hold. The line has bent. The next five years of AI development are where the question of whether it has broken gets answered, and Gilder has placed himself firmly on the side that says it will hold.
Why the Book Matters Now
Gaming AI was published five years ago. It is more relevant now than it was then.
The reason is that the questions it raises are the questions the entire AI industry is now forced to answer. What does it mean for a system to think. What is the economic value of human creativity if machines can simulate most of its outputs. What kinds of work survive automation, and what does the labor market look like on the other side. These were abstract questions in 2020. They are operational questions in 2026, and every company with a head of AI strategy is being paid to answer them.
Gilder’s answers come from a man who has spent five decades betting that the human mind is the scarcest resource in any economy, and who has been right about the direction of enough technology shifts to deserve a hearing on this one. The book is short, the argument is clear, and the conclusion is contestable. The contest is the point.
Read Gaming AI to understand the intellectual frame behind every appearance Gilder gave in 2026, and behind the investment theses that depend on AI moving from the cloud to the edge. The thesis is that minds win. The investment thesis is that the infrastructure to let them keep winning is where the capital flows. The first claim turns on a philosophical question about what thinking is. The second claim turns on whether the infrastructure buildout follows the trajectory Gilder’s framework predicts. Gilder has bet his career on both, and the rest of the Guru Files follows the arc.