AI manufacturing bear Gary Marcus called the AI capex boom “the biggest capitalization in history.” Goldman Sachs analyst Eric Sheridan reaches the opposite conclusion in his book “AI in a Bubble?” research package. Sheridan argues that this is not a cycle of hope and hype like 1999 but a cycle of scale and capitalization, with tangible income growth and extraordinary market momentum.
So, who is right? Jobs, pensions, and billions of dollars in the stock market, are at stake and affect all of us.
I focus on Amazon Web Services (AWS) as the most instructive window into the wider debate: it is the largest cloud enterprise, the one that discloses the cleanest revenue, and the one whose CEO has put the most clear defense on the table.
The chart below previews where this analysis resides: three tangible curves of AWS revenue, all corresponding to data through Q1 2026, each of which represents a different return on Amazon’s $200 billion spending plans this year. The disagreement between the bulls and the bears is a disagreement about where the curve is coming from.
Bulls say hyperscalers are funding this build with cash flow instead of debt, making the AI capex boom different from the history of telecom and train bubbles. Indeed, AWS grew by 28% last quarter, its fastest pace in 15 quarters, confirming that business demand for AI computing is real and accelerating.
Amazon CEO Andy Jassy has framed the company’s $200 billion capex plan as demand-driven rather than speculative, with strong returns expected on investments. Stanford professor Gilad Allon offers a more robust non-Wall-Street version of the same argument: AI build-out is funded mainly by cash-strapped incumbents rather than highly profitable speculative entrants, and high barriers to entry with chips, data centers, and power limitations are the kind of disparate build-outs that produce the classic bull run. Basically, the bull’s case is that the technology is real, the need is real, and too much caution is the type of error.
Conversely, the bears argue that AI capex calculations are based on assumptions that current operating numbers do not support.
Venture capitalist Tom Tunguz notes that Bank of America projects the issuance of hyperscaler loans of 175 billion dollars this year, six times the average of the previous five years – a sharp departure in the story funded by the bulls that rely on it. Protecting asset stability comes from Microsoft’s admission that $37.5 billion in the first quarter was allocated to short-term assets, especially GPUs that depreciate over five years instead of a thirty-year horizon for telecom or rail.
Apart from the question of the curve, the bears point to financial weaknesses that work without need: Oracle’s profit, Amazon’s sharp pivot of debt financing, and circular customer financing programs that include hyperscaler income in a small number of lab models whose income depends on large markets that remain open. Basically, the bear’s case is that the financial structure is changing, demand forecasts are weak, and being too aggressive causes a financial crisis.
If we return to our chart, the structure of the disagreement becomes concrete. The bull case assumes that the recent acceleration in AWS growth is the new normal and that growth rates continue to rise – generating approximately $66 billion in quarterly revenue by Q4 2027 and AWS quality returns to $200 billion capex.
The bear case assumes that the recent acceleration was a sustained move and that sequential dollar additions stabilize at around $2 trillion per quarter – generating around $52 billion in quarterly revenue and a welcome but disappointing return.
The case of the crisis is under the bear case: The demand for AI work is declining, and the GPU layer is no longer receiving enough money to recover its costs. The gap between a bull and a bear is not whether capex pays but how well it does.
Think back to the late 1990s, when phone companies laid more than 80 million miles of fiber-optic cable across the US to carry emerging Internet data traffic. It did not collapse because the drivers ran out of money. It collapsed because WorldCom told the market that Internet traffic was doubling every 100 days when the real rate was once a year. The predicted curve was closed, and money flowed accordingly.
By 2002, 85 to 95% of the fiber laid in the 1990s remained black, and nearly two billion dollars in market value had been removed. The demand eventually arrived – YouTube, streaming, the cloud – but it arrived ten years later, and the people who built it lost their shirts. The relevant question for AI is not whether the demand exists, which it clearly does, but whether it is growing fast enough to absorb $700 billion in annual capex.
The data that resolves the disagreement is 12 months out and will come in the form of average quarterly revenue. By Q1 2027, the difference between the bull and bear lanes is visible in the AWS data: at that time, AWS’s quarterly revenue will either accelerate to $40 billion, track against the low $40 billion, or show the first signs of decline.
None of those results are currently indisputable from the trajectory through Q1 2026, which is why hyperscalers can continue to raise debt and the market keeps buying it. Anyone who tells you for sure which curve will appear is selling something.
As for me, I just bought 12 months worth of popcorn.
[Editor’s note: GeekWire publishes guest opinion pieces representing a range of perspectives. The views expressed are those of the author.]
