While the US economy recorded just 1.5% real GDP growth in the second quarter of 2026, Cathie Wood, CEO of ARK Invest, has stunned the market with a bold forecast: the exponential expansion of artificial intelligence is pushing US real GDP growth toward double-digit territory. This assertion is not an isolated call but is grounded in data that traditional investors find hard to grasp—AI inference token usage has surged 25-fold within a year and continues to multiply at an exponential pace. Over the weekend, Tesla and SpaceX CEO Elon Musk publicly endorsed this view, declaring that "the AI wave has already begun." This AI macroeconomic narrative, ignited by Wood and crowned by Musk, is pushing capital markets toward a new pricing logic—and traditional enterprises may emerge as the biggest losers in this transformation.

The "25x Token Leap": The Transmission Logic from Micro to Macro

Wood wrote on X: "When I try to explain that inference token usage has grown 25x in a year, is doubling exponentially, is cascading through the economy like a waterfall, and will likely push real GDP growth to double digits per annum, many clients and companies simply can't compute this." Wood was responding to a post from Musk, in which he declared that "the AI wave has started." To help investors grasp the true meaning of "25x," Wood offered a simple quantification in a follow-up post: "Because they've never experienced 25x growth, most investors hear the number 25 and automatically convert it to 25%—which historically has been very good growth. But 25x is an investment concept from another dimension: 100 units of growth doesn't become 125, it skyrockets to 2,500!"

This explosive growth in micro-level data is rooted in the cliff-like decline of AI inference costs. According to ARK's "Big Ideas 2026" report released earlier this year, unit inference costs have fallen more than 90%, directly stimulating rapid enterprise adoption of AI computing power. On certain benchmarks, annual inference cost reductions reach as high as 99%, while training costs decline at roughly 75% per year. From collapsing costs to exploding demand, and then to macroeconomic growth, ARK's logic chain is clear and aggressive: cheaper AI inference → more token consumption → greater AI adoption across economic activities → productivity surge. Wood previously predicted that AI-driven productivity gains could push real GDP growth to nearly 5%—the "Goldilocks" level—in 2026, and she has now revised that upward to double digits. ARK's model assumes 5% to 7% productivity growth, approximately 1% labor force growth, and an inflation range of negative 2% to positive 1%, together supporting nominal GDP growth of 6% to 8%. Wood emphasizes that AI-driven productivity growth tends to lower inflation rather than raise it—a view that stands in stark contrast to the US second-quarter PCE inflation reading of 3.7%.

However, a vast gulf separates Wood's aggressive forecast from the reality of US economic data. Preliminary figures released by the US Commerce Department on July 30 show that real GDP annualized growth slowed to 1.5% in the second quarter of 2026, below both the first quarter's 2.1% and the 2.1% economists had broadly anticipated.

A Stark Warning: Traditional Enterprises Face a "Capital Black Hole" as AI Drains the Economy's Resources

If Wood's GDP forecast is the "offensive end" of this narrative, then Brett Winton, ARK's Chief Futurist, delivers the "defensive end" warning—and its impact may far exceed what most people imagine. Winton posted on X a judgment capable of keeping any traditional industry CEO awake at night: "The fast payback periods and extremely high internal rates of return (IRR) on AI infrastructure, even at massive scale, will push up the broader cost of capital enough to drive many traditional enterprises into the abyss—even those not directly exposed to AI disruption."

Winton's logic chain is sharp and lethal: the return on AI infrastructure investment is "absurdly high," with increasingly shorter payback periods. As long as such ultra-high returns persist, capital will keep flowing toward GPUs, data centers, and AI companies like sharks drawn to blood. The end result: many traditional companies may not even need to face AI competition head-on—capital will leave them first. As financing gets more expensive and investors demand higher returns, some companies that could otherwise operate normally may be gradually squeezed out of the market.

Wood then added another set of data points to support this macro narrative: frontier AI lab revenues have grown 5 to 10 times within six months to a year; some mature companies benefiting from the AI dividend have seen revenue growth re-accelerate from 25% to 30% to 40% or more. She described many investors as "deer caught in headlights"—they can see the change barreling toward them but don't yet understand what it truly means.

JPMorgan's Quantitative Corroboration

This logic is being validated by macro data. JPMorgan predicted in June that AI-related debt financing could reach $4.1 trillion by 2030, with total AI capital expenditure climbing to $5.5 trillion over the same period. Loans cover on average 85% of total project costs, prompting companies to leverage "every capital market" to meet growth financing needs. Specifically, JPMorgan forecasts that the high-grade bond market will provide $2.1 trillion in financing for data centers over the next five years, the leveraged finance market $350 billion, with another $1 trillion coming from internal cash flow, $400 billion from incremental equity capital, and $300 billion from structured products. The report notes that hyperscale data center operators continue to maintain "astonishing profitability," with cash flows projected to exceed $900 billion by 2027. Meanwhile, due to surging memory costs at Samsung Electronics, SK Hynix, and Micron Technology, Nvidia has already issued price adjustment notices to core customers including Microsoft, Google, and Oracle, with server prices equipped with its AI chips set to rise by more than 15% on shipments beginning in early 2027.

When "Capital Siphoning" Becomes the Defining Variable of the AI Era, a Paradigm Shift in Capital Allocation Begins

When Wood wrote on X that "many clients and companies simply can't compute this," she may have been referring not only to the market's misjudgment of AI technology diffusion speed, but also to the neglect of a deeper trend: AI is systematically reshaping global capital allocation logic. The true value of this "AI GDP conversation" between Wood and Musk across X lies not in the accuracy of their predictions, but in revealing a structural change that is happening now yet rarely acknowledged: AI is becoming a massive capital-siphoning machine. Traditional enterprises unrelated to AI are not dying because their technology lags behind—but because the money is being drained away by AI, slowly suffocating them. This is a silent "de-industrialization" process driven by differential capital returns. As Winton noted, we used to debate which jobs AI would take and whose income would be affected. The next question may be more fundamental: when the AI industry persistently offers capital returns far exceeding traditional sectors, will it draw all the money in the economy toward itself? If the answer is yes, then the real sign of AI beginning to disrupt traditional industries may not be the release of a more powerful model—but capital voting with its feet. Wood's 25x token growth and double-digit GDP forecast are the early echoes of this paradigm shift. These numbers together paint a picture that transcends the realm of "tech stock ups and downs." They point to a more fundamental question: when the AI industry's capital returns persistently outpace traditional sectors over the long term, the resource allocation logic of the entire economy will be completely rewritten.