This article appears in the April 2025 issue of The American Prospect magazine. Subscribe here.

The week of Donald Trump’s inauguration, Sam Altman, the CEO of OpenAI, stood tall next to the president as he made a dramatic announcement: the launch of Project Stargate, a $500 billion supercluster in the rolling plains of Texas that would run OpenAI’s massive artificial-intelligence models. Befitting its name, Stargate would dwarf most megaprojects in human history. Even the $100 billion that Altman promised would be deployed “immediately” would be much more expensive than the Manhattan Project ($30 billion in current dollars) and the COVID vaccine’s Operation Warp Speed ($18 billion), rivaling the multiyear construction of the Interstate Highway System ($114 billion). OpenAI would have all the computing infrastructure it needed to complete its ultimate goal of building humanity’s last invention: artificial general intelligence (AGI).

Art for this story was created with Midjourney 6.1, an AI image generator.

But the reaction to Stargate was muted as Silicon Valley had turned its attention west. A new generative AI model called DeepSeek R1, released by the Chinese hedge fund High-Flyer, sent a threatening tremor through the balance sheets and investment portfolios of the tech industry. DeepSeek’s latest version, allegedly trained for just $6 million (though this has been contested), matched the performance of OpenAI’s flagship reasoning model o1 at 95 percent lower cost. R1 even learned o1 reasoning techniques, OpenAI’s much-hyped “secret sauce” to allow it to maintain a wide technical lead over other models. Best of all, R1 is open-source down to the model weights, so anyone can download and modify the details of the model themselves for free.

It’s an existential threat to OpenAI’s business model, which depends on using its technical lead to sell the most expensive subscriptions in the industry. It also threatens to pop a speculative bubble around generative AI inflated by the Silicon Valley hype machine, with hundreds of billions at stake.

Venture capital (VC) funds, drunk on a decade of “growth at all costs,” have poured about $200 billion into generative AI. Making matters worse, the stock market’s bull run is deeply dependent on the growth of the Big Tech companies fueling the AI bubble. In 2023, 71 percent of the total gains in the S&P 500 were attributable to the “Magnificent Seven”—Apple, Nvidia, Tesla, Alphabet, Meta, Amazon, and Microsoft—all of which are among the biggest spenders on AI. Just four—Microsoft, Alphabet, Amazon, and Meta—combined for $246 billion of capital expenditure in 2024 to support the AI build-out. Goldman Sachs expects Big Tech to spend over $1 trillion on chips and data centers to power AI over the next five years. Yet OpenAI, the current market leader, expects to lose $5 billion this year, and its annual losses to swell to $11 billion by 2026. If the AI bubble bursts, it not only threatens to wipe out VC firms in the Valley but also blow a gaping hole in the public markets and cause an economy-wide meltdown.

OpenAI’s Ever-Increasing Costs

The basic problem facing Silicon Valley today is, ironically, one of growth. There are no more digital frontiers to conquer. The young, pioneering upstarts—Facebook, Google, Amazon—that struck out toward the digital wilderness are now the monopolists, constraining growth with onerous rentier fees they can charge because of their market-making size. The software industry’s spectacular returns from the launch of the internet in the ’90s to the end of the 2010s would never come back, but venture capitalists still chased the chance to invest in the next Facebook or Google. This has led to what AI critic Ed Zitron calls the “rot economy,” in which VCs overhype a series of digital technologies—the blockchain, then cryptocurrencies, then NFTs, and then the metaverse—promising the limitless growth of the early internet companies. According to Zitron, each of these innovations failed to either transform existing industries or become sustainable industries themselves, because the business case at the heart of these technologies was rotten, pushed forward by wasteful, bloated venture investments still selling an endless digital frontier of growth that no longer existed. Enter AGI, the proposed creation of an AI with an intelligence that dwarfs any single person’s and possibly the collective intelligence of humanity. Once AGI is built, we can easily solve many of the toughest challenges facing humanity: climate change, cancer, new net-zero energy sources.

And no company has pushed the coming of AGI more than OpenAI, which has ridden the hype to incredible heights since its release of generative chatbot ChatGPT. Last year, OpenAI completed a blockbuster funding round, raising $6.6 billion at a valuation of $157 billion, making it the third most valuable startup in the world at the time after SpaceX and ByteDance, TikTok’s parent company. OpenAI, which released ChatGPT in November 2022, now sees 250 million weekly active users and about 11 million paying subscribers for its AI tools. The startup’s monthly revenue hit $300 million in August, up more than 1,700 percent since the start of 2023, and it expects to clear $3.7 billion for the year. By all accounts, this is another world-changing startup on a meteoric rise. Yet take a deeper look at OpenAI’s financial situation and expected future growth, and cracks begin to show.

To start, OpenAI is burning money at an impressive but unsustainable pace. The latest funding round is its third in the last two years, atypical for a startup, that also included a $4 billion revolving line of credit—a loan on tap, essentially—on top of the $6.6 billion of equity, revealing an insatiable need for investor cash to survive. Despite $3.7 billion in sales this year, OpenAI expects to lose $5 billion due to the stratospheric costs of building and running generative AI models, which includes $4 billion in cloud computing to run their AI models, $3 billion in computing to train the next generation of models, and $1.5 billion for its staff. According to its own numbers, OpenAI loses $2 for every $1 it makes, a red flag for the sustainability of any business. Worse, these costs are expected to increase as ChatGPT gains users and OpenAI seeks to upgrade its foundation model from GPT-4 to GPT-5 sometime in the next six months.

Financial documents reviewed by The Information confirm this trajectory as the startup predicts its annual losses will hit $14 billion by 2026. Further, OpenAI sees $100 billion in annual revenue—a number that would rival Nestlé and Target’s returns—as the point at which it will finally break even. For comparison, Google’s parent company, Alphabet, only cleared $100 billion in sales in 2021, 23 years after its founding, yet boasted a portfolio of money-making products, including Google Search, the Android operating system, Gmail, and cloud computing.

OpenAI is deeply dependent on hypothetical breakthroughs from future models that unlock more capabilities to boost its subscription price and grow its user base. Its GPT-5 class models and beyond must pull godlike capacities for AI out of the algorithmic ether to create a user base of hundreds of millions of paid subscribers. Yet, with the release of the open-source DeepSeek R1 model earlier this month, OpenAI has no moat for its increasingly expensive products. The R1 matched its performance across math, chemistry, and coding tasks, independently learned OpenAI’s reasoning techniques, and can be downloaded, modified, and deployed for free. Why would people continue to pay $20, let alone the $200 tier OpenAI reserves for its latest, greatest models, rather than use something that can deliver the same performance at a 95 percent lower price?

Venture capital funds, drunk on a decade of “growth at all costs,” have poured about $200 billion into generative AI.

Silicon Valley Is All In on AI

Wall Street asked itself the same question after the release of DeepSeek R1 and panicked, wiping more than 15 percent ($600 billion) off Nvidia’s stock price, the largest single-day loss for a company ever. And that’s not the only bad sign Altman received about OpenAI’s future. OpenAI is in talks again to raise more money (less than a year after raising $10 billion) at a proposed $340 billion valuation. In most cases, a startup doubling its valuation would be great news. But for OpenAI, the money may make things worse. It signals a desperate need for cash and puts more pressure on a company that today loses $2 for every dollar it makes. As Zitron pointed out, at $340 billion, few companies have the liquidity to acquire OpenAI, and public investors expect strong returns and profitability to justify an IPO anywhere near that price. Plus, the latest round of funding is being led by Masayoshi Son, a billionaire investor known more for losing money than making it. Given Son’s Vision Fund’s disastrous investing record, Zitron said, it’s as bearish a signal as you could find. Hanging over all this for OpenAI is the fact that Microsoft’s investments in the company, which run north of $10 billion, are not standard equity investments but “profit participation units” that will convert to debt in a year and a half.

It’s not just OpenAI that’s burning through billions. Silicon Valley has hyped AI as the next internet or iPhone, and has invested like it cannot afford to miss out on the next big tech revolution. In 2021, with the last gasp of zero-interest-rate loans paired with trillions in COVID relief spending, venture capitalists poured a record $78.5 billion into the AI space. And, despite a broader slowdown in venture activity, the second quarter of 2024 set the record for quarterly venture investing in AI at $23.3 billion. In fact, 33 percent of VC portfolios are committed to AI, another worrying sign of concentration.

Even so, Big Tech companies are the biggest spenders on AI. While VCs dropped approximately $200 billion into AI between 2021 and 2024, Big Tech is on pace to surpass that amount this year alone. According to Goldman Sachs research, cloud computing giants are expected to plow over $1 trillion over the next five years into graphics processing units (GPUs) and to build data centers to power generative AI. AI is an expensive technology like few before it.

All those racks of GPUs and supercluster data centers need power, and the power industry is also embarking on a once-in-a-generation investment spree to keep up. The scale of expected data centers to power generative AI is difficult to wrap your head around. Oracle recently announced plans to build a gigawatt-scale data center just for AI, powered by a trio of nuclear reactors, while OpenAI pitched the White House on the necessity of five-gigawatt data centers, which would be enough power for about three million homes consumed by a single AI data center. A recent report from McKinsey expects the electricity going to fuel AI data centers will triple from 3 to 4 percent of the country’s electricity to 11 to 12 percent by 2030. The power industry typically grows 2 to 3 percent a year, far too little to meet the predicted jump in demand. McKinsey estimates that power utilities would have to spend $500 billion on top of their planned capital expenditure to keep up with AI needs. If true, this presents a serious bottleneck for not just OpenAI but the expected growth of the AI industry.

Where’s the Money, Lebowski?

Between VCs, Big Tech, and power utilities, the bill for generative AI comes out to close to $2 trillion in spending over the next five years alone. Adding all this up, some are starting to question the economic fundamentals of generative AI. Jim Covello, head of global equity research at Goldman Sachs, doubts the technology can recoup what’s been invested as, unlike the internet, it fails to solve complex business problems at a lower cost than what’s available today. Plus, he argues, the most expensive inputs for generative AI, GPUs and energy, are unlikely to decline meaningfully for the tech industry over time, given how far demand outstrips supply for both. While AI-fueled coding could definitely boost productivity, it’s hard to see how it could become a multitrillion-dollar industry.

Surveys confirm that for many workers, AI tools like ChatGPT reduce their productivity by increasing the volume of content and steps needed to complete a given task, and by frequently introducing errors that have to be checked and corrected. A study by Uplevel Data Labs tracked 800 software engineers using Copilot on GitHub and found no measurable increase in coding productivity, despite this exact use case being the one pointed to the most by AI companies. And even productivity gains may come at a cost: Microsoft researchers concluded that workers became more productive using generative AI tools but their critical thinking skills declined, presumably because they were offloading the thinking to AI. Looking past the hype, the business case for generative AI two years after the stunning success of ChatGPT appears weaker by the day.

Even worse, as AI expert Gary Marcus pointed out, DeepSeek’s R1 model spells serious trouble for OpenAI and the cloud giants. The only way OpenAI could hope to recoup the billions it was spending on GPUs to train bigger and bigger models was to maintain a large enough technical lead over other AI companies to justify charging up to $200 for paid subscriptions to its models. That lead just vaporized and was given to the entire industry for free. In response, Altman has already twice cut the prices of his subscriptions in an effort to stay competitive. But without millions of paid subscriptions, it’s difficult to see the pathway to profitability for a company that loses $2 for every $1 it brings in and expects costs to continue to grow approximately tenfold in five years. OpenAI has set $100 billion as its break-even point, which would require it to increase its revenue by a factor of 25 in just five years, an incredible feat of scale that its current business model does not justify.

OpenAI is, however, the perfect kind of growth-at-all-costs story investors need to think still exists—capable of not only achieving Meta- or Amazon-like growth again, but becoming an indispensable part of growth and innovation in every industry in the future, too. No industry could escape, and software would close its jaws around the world, finally, as Marc Andreessen predicted in 2011.

For his part, Gary Marcus has taken to calling OpenAI the WeWork of AI—WeWork, of course, is the poster boy for wasteful, nearly fraudulent, growth-at-all-costs investing that led to a spectacular downfall. Marcus is so confident current approaches cannot take us to the promised land of AGI that he bet Anthropic CEO Dario Amodei $100,000 that AGI would not be achieved by the end of 2027. Without AGI, the valuations of leading AI startups like OpenAI ($340 billion) and Anthropic ($61.5 billion) stop making sense. If GPUs are no longer the most capital-efficient or effective way to build better AI models, then the expected AI computing “supercycle” that the hundreds of billions in capital expenditure is premised on never arrives. Instead, the underlying asset bubble of a multitrillion-dollar bet on GPUs as the necessary component to an internet-like era of growth vanishes into thin air.

For many workers, AI tools reduce their productivity by increasing the volume of steps needed to complete a given task.

Just How Big Will the Blast Be?

OpenAI’s incredible burn rate, the trillions in capital expenditure by cloud giants and utilities to build out the infrastructure necessary to support AI, the supply bottlenecks ahead from the power and semiconductor industries, and the questionable economic gains from these tools all point to a generative AI bubble. Should the bubble burst, startups and venture funds alike face possible extinction, and a big enough drop from the Magnificent Seven could spark skittish markets to panic, leading to wider economic contagion, given how dependent on the growth of the top technology companies the public markets have become.

In 2024, the Magnificent Seven were responsible for the lion’s share of the growth of the S&P 500, with the returns of the other 493 companies flat. When Nvidia hit its peak valuation of $3 trillion over the summer, just five of the seven—Microsoft, Apple, Nvidia, Alphabet, and Amazon—accounted for 29 percent of the total index’s value, surpassing the concentration of the five top technology companies just before the dot-com crash. Nvidia has been on an incredible bull run over the last five years, its shares gaining a dizzying 4,300 percent, reminiscent of how network equipment maker Cisco grew about 4,500 percent in the five years leading up to its peak just before the dot-com crash in 2000.

Nvidia and the other Magnificent Seven members are in a codependent relationship when it comes to AI hype. They are Nvidia’s biggest customers, feeding the bull run by pushing demand for GPUs beyond even what chipmaker TSMC can supply. At the moment, Nvidia can pass those prices on to their customers, the only clusters big enough for AI computing. But should demand for AI fall, all seven will tumble with it.

For the tech industry, DeepSeek is a threat to its incredible bull run because it proved three things. First, frontier AI models could be trained much more cheaply and efficiently than the current Silicon Valley approach of building massive models requiring hundreds of thousands of GPUs to train. From a capital perspective, the U.S. strategy is wasteful, relying on at least ten times the investment to make similar model progress. Second, DeepSeek showed you could train a state-of-the-art model without the latest GPUs, calling into question the current demand for the latest GPUs that is so hot customers have been facing delays of six months to a year to get their hands on them. Finally, the high valuations of leading AI startups depend on a technical lead in their models to charge prices anywhere near what they need to recoup their computing costs, but that technical lead, enabled by a combination of closed-source models, billions in capital expenditure, and export controls blocking Chinese companies like DeepSeek from accessing the latest GPUs, is gone. Should demand for GPUs fall or even not hit the exponential increases the billions invested are betting on, the bubble will pop.

Given the stock market’s dependence on tech companies for growth, the trigger may not come from the AI industry itself, but any pullback in spending will crater the current trajectory of the AI industry. Many potential triggers abound: a crypto crash; President Trump’s trade wars with Canada, Mexico, and China; the stated goal to cut more than $1 trillion of government spending by Elon Musk’s Department of Government Efficiency; or a Chinese invasion of Taiwan, where nearly 70 percent of the world’s advanced computer chips are manufactured. You can tell Wall Street is worried about a bubble, because Nvidia is hit the hardest by any bearish AI news, and even when the market panic has nothing to do with the tech industry, like when the Japan currency trade happened last summer, the Magnificent Seven suffer punishing losses.

The AI bubble wobbles more precariously by the day. Some bubbles, like that of the dot-com crash, end up being positive in the long run, despite the short-term economic pain of it bursting. But some, like the 2008 housing bubble, leave permanent scars on the economy and can knock an entire industry off its growth trajectory for years. To date, the U.S. housing industry has not recovered to pre-2008 growth trend lines, a major contributor to the housing crisis gripping the U.S. That is the fire that the tech industry is playing with today.

This is not the Silicon Valley of lore. Venture investors, for all their tech manifestos celebrating “little tech” and entrepreneurship, have come to resemble more traditional financial firms, raising money from pension funds, hedge funds, and sovereign wealth funds. Silicon Valley has gone corporate and managerial; even private equity invests in the Valley today. The fusion of venture capital and Wall Street threatens to bring the unbridled speculation of unregulated finance and the breathless tech industry hype together in a single, massive bubble. Inimical to the old ethos of the Valley and emblematic of a bloated, rotten investing strategy, in Silicon Valley now the money chases founders rather than founders chasing money. Maybe, after the fallout of the AI bubble is felt and the sun sets on Silicon Valley for a bit, the tech world can do a hard reset and return to its more innovative days again.