Epoch AI Researcher: "Follow the Money" to Understand True AI Progress
Revenue growth at leading AI companies provides a far more objective measure of artificial intelligence progress than flashy capability demonstrations or academic benchmarks, according to a prominent AI forecaster at Epoch AI.
Ege Erdil, a researcher specializing in AI timeline forecasting, argues that investors and policymakers should focus on commercial success rather than headline-grabbing capability demonstrations when assessing the field's trajectory, writes End of Miles.
Why revenue tells the real story
"The most informative thing for me has been the fact that I was too pessimistic about revenue growth," Erdil explained during a recent discussion on AI development timelines. "I expected revenue growth to slow down as the labs reached higher levels of revenue, and that doesn't seem to be happening."
The AI forecaster pointed to OpenAI's financial trajectory as a clear indicator of faster-than-expected commercial adoption. "I'm expecting revenue this year for OpenAI to be around 12 billion or something, so it's like three times more than last year. And last year was maybe three times more than the year before that. I think that is pretty impressive."
"OpenAI is forecasting that they're going to have hundred billion in revenue by 2029... that's their own internal forecast. And now that just doesn't look implausible to me." Ege Erdil, Epoch AI
The AI researcher had previously predicted such revenue levels would take a full decade to achieve, not the five to six years now projected by OpenAI itself.
Less flash, more cash
This focus on commercial success stands in stark contrast to the attention typically given to capability demonstrations like mathematical reasoning, complex problem-solving, or programming tournaments.
"They pay more attention to the reasoning models which do math and complex reasoning and programming," the forecasting expert noted. "They look at the Codeforces ELO of the models and how quickly it's progressed, while for me, it's much more relevant that revenue growth has been continuing at a speed faster than I expected."
This distinction matters because capability demonstrations may not directly translate to economic value. Erdil argues that if we're interested in predicting when AI will truly transform the economy, we should look at what customers are actually willing to pay for, not what impresses researchers.
What this means for understanding AI's future
The Epoch researcher suggests this focus on revenue creates a more grounded approach to understanding AI's development trajectory. While mathematical reasoning models may generate excitement in research communities, their immediate commercial impact remains limited compared to general-purpose models that solve everyday business problems.
"We haven't seen as rapid progress on the capabilities that would actually unlock a lot of economic value," the AI forecaster explained. "I think it's much more important to have agency and executive function and ability to adapt plans to changing circumstances."
"If you look at our economy, we don't pay mathematicians that much. They're just a very small fraction of the economy. Clearly, even if you automated that, how much value would that create?" Erdil
The economist-turned-AI-researcher points out that mathematics represents just one input in technological progress, alongside countless other bottlenecks. Simply automating mathematical reasoning isn't enough to drive massive economic growth.
The market is voting with its dollars
This revenue-focused approach provides a market-based reality check on AI hype. While researchers might celebrate when a model solves complex mathematical proofs, businesses are primarily investing in AI that can perform practical tasks with sufficient reliability.
"I don't feel like we have a clear trend that we can extrapolate where models are just becoming better at these kinds of tasks," Erdil observed regarding agentic capabilities that would enable AI to function more independently in real-world environments.
The revenue growth at companies like OpenAI suggests they've found product-market fit despite these limitations, indicating that current capabilities already deliver sufficient value to drive rapid commercial adoption.
"I would now say maybe 30 years for when we're going to have an AI system that could do any job that could in principle be done remotely. I said 40 years previously, so I would be more bullish on that than I used to be." The Epoch researcher
This subtle shift in timeline forecasting represents a significant update based on observed commercial adoption rather than technological demonstrations alone.
The bottom line for decision-makers
For investors, policymakers, and business leaders trying to navigate AI's uncertain future, Erdil's perspective offers a pragmatic framework: follow the money, not the demos.
The Epoch forecaster's emphasis on revenue growth over capability benchmarks suggests a more reliable approach to understanding AI's economic impact trajectory. Despite remaining relatively conservative on long-term transformative potential, his recognition of faster-than-expected commercial adoption indicates meaningful progress in AI's practical applications.
As AI companies continue scaling their revenue at rates exceeding most analyst predictions, this commercial success may prove to be the most reliable indicator of when and how artificial intelligence will reshape the global economy—regardless of how many mathematical proofs the latest models can solve.
