The two top players are optimistic about the future of AI! Musk chanted “The big wave of AI has arrived,” and Sister Mu made a high-profile prediction that the US GDP will experience double-digit growth

The Zhitong Finance App learned that when the US economy recorded only 1.5% real GDP growth in the second quarter of 2026, Cathie Wood, CEO of ARK Invest, known as “Sister Mu”, issued an astonishing prediction: the exponential expansion of artificial intelligence is pushing the real GDP growth rate of the US into a double-digit range.

This assertion is not an isolated statement, but is based on a set of data that is difficult for traditional investors to understand: AI inference token usage has increased 25 times in a year and is still doubling at an exponential rate. Over the weekend, Tesla and SpaceX CEO Elon Musk publicly endorsed this judgment, saying bluntly that “the big wave of AI has begun.” This AI macro-narrative, ignited by “Sister Mu Tou” and crowned by Musk herself, is pushing the capital market towards a new pricing logic — and traditional companies are likely to be the biggest losers in this transformation.

Token's “25x transition”: transmission logic from micro to macro

Wood wrote on the X platform: “Many customers and companies simply couldn't understand when I tried to explain that token usage has increased 25 times in a year and is multiplying exponentially, sweeping the entire economy like a waterfall, and is likely to push real GDP growth to double digits every year.”

Wood was responding to a post by Musk in which he declared that “the wave of artificial intelligence has begun.”

To help investors understand the true meaning of “25 times,” Wood did a popular quantification in a follow-up post: “Since we've never experienced a 25-fold increase, most investors automatically convert it to 25% when they hear the number 25 — this is indeed a great increase in history. But 25x is an investment concept in another dimension: the increase of 100 units did not become 125, but soared to 2,500!”

This explosive growth in microdata is rooted in a cliff-style decline in AI inference costs. According to the “Big Ideas 2026” report released by ARK at the beginning of the year, AI unit inference costs have been reduced by more than 90%, directly stimulating the rapid adoption of AI computing power by enterprises. In some benchmarks, inference costs dropped by as much as 99% per year, while training costs dropped at a rate of about 75% per year.

From cost collapse to demand explosion to macroeconomic growth, ARK's logical chain is clear and aggressive: cheaper AI reasoning → more token consumption → AI transformation of more economic activity → productivity jump. Wood previously predicted that AI-driven productivity growth could push the real GDP growth rate to close to 5% of the “Goldilocks” level in 2026, but now she has further raised it to double digits.

ARK's model assumes 5% to 7% productivity growth, labor force growth of about 1%, and an inflation range of negative 2% to positive 1%, which together support nominal GDP growth of 6% to 8%. Wood emphasized that AI-driven productivity growth tends to reduce inflation rather than drive up inflation — a judgment in stark contrast to the US PCE inflation data of 3.7% in the second quarter.

However, there is a huge gap between Wood's aggressive predictions and actual data on the US economy. According to preliminary data released by the US Department of Commerce on July 30, the annualized growth rate of US real GDP in the second quarter of 2026 has slowed to 1.5%, not only lower than 2.1% in the first quarter, but also far lower than the 2.1% generally predicted by economists.

Brutal warning: traditional enterprises are facing a “capital black hole”, and AI is absorbing capital from the entire economy

If Wood's GDP forecast is the “offensive side” of this story, then ARK's chief futurist Brett Winton's warning is the “defensive side” of this change — and it's probably far more lethal than most people think.

Winton threw a judgment on X that would make any traditional industry CEO sleepy: “The rapid payback period and extremely high internal rate of return (IRR) of AI infrastructure, even on a large scale, will drive up a wider range of capital costs, enough to push many traditional enterprises into the abyss — even those that are not clearly directly impacted by AI.”

Winton's logical chain is sophisticated and deadly: the return on investment in AI infrastructure is “ridiculously high”, and the payback period is getting shorter and shorter. As long as this high return continues to exist, capital will continue to flock to GPU, data centers, and AI companies like a shark that smells of blood. The end result: many traditional businesses don't even have to actually face AI competition—capital leaves them first. Financing is getting more expensive, and investors are demanding higher returns — some companies that can still operate normally may slowly be squeezed out of the market.

Wood then added another set of data to support this macro-narrative: revenue from cutting-edge AI labs increased 5 to 10 times in half a year to a year; some mature companies benefiting from AI dividends have re-accelerated their revenue growth from 25% to 30% to over 40%. She described many investors as “like deer shining in car lights” — they've seen the changes come through, but haven't understood what it really means.

Quantitative evidence from J.P. Morgan

This logic is being confirmed by macroeconomic data. J.P. Morgan Chase predicted in June that AI-related debt financing could reach 4.1 trillion US dollars by 2030, and total AI capital expenditure will rise to 5.5 trillion US dollars during the same period. Loans cover an average of 85% of the total project cost, driving companies to use “every capital market” to meet growth financing needs.

Specifically, J.P. Morgan predicts that in the next five years, the high-rated bond market will provide 2.1 trillion US dollars of financing for data centers, 350 billion US dollars for the leveraged financing market, another 1 trillion US dollars from internal cash flow, 400 billion US dollars from incremental equity capital, and 300 billion US dollars from structured products. According to the report, hyperscale data center operators are still maintaining “amazing profitability,” and the cash flow of these companies is expected to exceed 900 billion US dollars by 2027.

Meanwhile, due to soaring memory costs for Samsung Electronics, SK Hynix, and Micron Technology, Nvidia has issued price adjustment notices to core customers such as Microsoft, Google, and Oracle. The price of servers equipped with its AI chips will generally rise by more than 15%, effective from early 2027.

When “capital siphon” became the biggest variable in the AI era, the “capital flow” paradigm shift quietly began

When Wood wrote “many customers and companies don't understand” on the X platform, she probably meant not only the market's misjudgment of the speed at which AI technology is spreading, but also ignoring the deep trend that AI is systematically changing the logic of global capital allocation.

The real value of this “AI GDP conversation” between Wood and Musk, which spans the X platform, is not the accuracy of predictions, but rather reveals a structural change that is taking place but is rarely being confronted with: AI is becoming a huge capital siphon machine. Those traditional companies that have nothing to do with AI are not dying because of poor technology — they are slowly being suffocated by money being sucked away by AI. This is a silent “de-industrialization” process driven by differences in return on capital.

As Winton said, in the past, we were always discussing whose jobs and whose income AI would rob. The next question is probably more fundamental: when the AI industry provides a return on capital far higher than traditional industries for a long time, will it absorb all of the money from the entire economy? If the answer is yes, then the sign that AI is actually beginning to impact traditional industries is probably not that it has created a stronger model — but rather that capital has begun to vote with its feet. Wood's 25-fold token growth and double-digit GDP forecasts are an early echo of this paradigm shift.

The picture these numbers together depict far exceeds the category of “rising and falling technology stocks.” It points to a more fundamental problem: when the return on capital of the AI industry is much higher than that of traditional industries for a long time, the entire economy's resource allocation logic will be completely rewritten.