Insiders warn of a 'vertical wall' in compute demand threatening OpenAI and rivals

The AI boom has a very physical problem. Models need chips, power, cooling, land, fiber, and long-term cloud contracts — and all of that gets expensive before the revenue shows up. That gap is what people inside the industry mean by a “vertical wall” in compute demand: spending rises almost straight up, while the business case still has to prove it can keep pace. The story snapped into focus on April 28, when reports that OpenAI had missed internal growth targets hit the market and dragged down a whole ring of partners around it. (forbes.com) ### What is the “vertical wall”? Basically, it is the point where AI demand stops looking like a software scaling problem and starts looking like an infrastructure financing problem. Training and serving bigger models means reserving massive compute years in advance, not just buying more servers next quarter. Once companies sign those deals, the costs become much less flexible than the hype makes them sound. (cnbc.com) ### Why is OpenAI at the center? Because OpenAI has become the anchor tenant for a lot of the industry’s biggest buildouts. It told investors in February that it now expects roughly $600 billion in total compute spend by 2030, down from the much looser $1.4 trillion infrastructure talk Sam Altman had floated earlier. The same investor pitch tied that spending to a very steep revenue ramp — from $13.1 billion in 2025 to more than $280 billion in 2030. (cnbc.com) ### Where does the huge headline number come from? It is not one invoice. It is a stack. Stargate alone was announced as a project targeting up to $500 billion of AI infrastructure in the U.S. OpenAI and Oracle later said their partnership exceeded $300 billion over five years, tied to as much as 4.5 additional gigawatts of capacity. Open(cnbc.com) next three years. (openai.com) ### So what changed this week? The market got reminded that these commitments depend on growth, not just excitement. The Wall Street Journal reporting, echoed by Reuters and others, said OpenAI missed several monthly revenue targets and fell short of an internal goal of 1 billion weekly ChatGPT users by the end of 2025. It also said CFO Sarah Friar had warned colleagues that, if growth does(openai.com)ontracts. (forbes.com) ### Didn’t OpenAI push back? Yes — hard. OpenAI called the report “clickbait” and said the business is “firing on all cylinders,” with strong consumer, enterprise, and developer growth. But the pushback did not stop investors from selling first and asking questions later. Oracle, CoreWeave, SoftBank, Nvidia, Microsoft, and Broadcom all traded lower after the report landed. (bloomberg.com) ### Why does this hit Oracle and others so fast? Because they are not just bystanders. Oracle is building a big chunk of the physical capacity OpenAI wants to consume. CoreWeave is another major compute supplier. Nvidia and Broadcom sit further upstream, but they still benefit if these giant clusters keep getting built. If Op(bloomberg.com) the contractors already ordered the steel. (openai.com) ### Is this a demand collapse? Not really. The catch is that demand for AI is still enormous. OpenAI said ChatGPT had more than 900 million weekly active users, and the company just raised capital at an $852 billion post-money valuation. The problem is narrower and more brutal: even huge demand may not be enough if the infrastructure curve rises faster than monetization. (cnbc.com)-around-600-billion-by-2030.html)) ### What’s the bottom line? The AI race used to look constrained by chip supply. Now it also looks constrained by who can safely underwrite years of compute before the cash flow is there. If the growth keeps compounding, these deals look visionary. If it does not, the “vertical wall” stops being a metaphor and turns into a balance-sheet problem. (cnbc.com)