How the coming AI-related IPOs could puncture the AI bubble
-
Click here to listen to this article - Share via
See more from the L.A. Times in Google Search. Set us as preferred
There’s an old gambler’s aphorism beloved by wise souls from Amarillo Slim to Warren Buffett that warns: “If you look around the poker table and can’t identify the mark, it’s you.”
Investors might be well advised to keep this thought in mind over the coming months. That’s because mega-initial public offerings from the widely followed artificial intelligence firms OpenAI and Anthropic are lined up for takeoff.
Public enthusiasm for these IPOs has been stoked by relentless PR about how AI is certain to change our lives, as well as by a gain in the first couple of trading days after last week’s SpaceX IPO, which was goosed by the Elon Musk company’s claim that it would ride the AI wave into the limitless future.
New issues have special salesmanship behind them, which calls therefore for a special degree of sales resistance.
— Value investing guru Benjamin Graham on IPOs (in 1949)
(Never mind that the gain largely evaporated Monday, when SpaceX shares fell to $154.60 — well below their closing price of $160.95 on their first day of trading on June 12.)
There are many reasons for the average investor to look askance at initial public offerings generically; more on those in a moment. But some are specific to the AI field, especially just now. The rush to public stock listings suggests that AI insiders sense that we are approaching peak AI — more specifically, that the AI bubble is about to burst.
To be fair, concerns about whether AI even has been a bubble have percolated for months. I joined that camp almost a year ago, after OpenAI’s then-most recent chatbot, GPT-5, turned out to be an underwhelming bust.
Since then, however, signs that AI has been oversold as a world-changing technology have proliferated. Let’s take a look.
First and foremost, many businesses that went all-in on AI have discovered that the vaunted cost-savings its promoters promised have been chimerical at best.
Some that pared their human workforces in favor of AI bots have had to rehire fired workers, according to Forrester Research. More than half of the firms surveyed by the British software consultancy OrgVue last year acknowledged that the firings were mistakes. AI critic Will Lockett dubs this the “AI boomerang effect.”
Local opposition has blocked or delayed more than a dozen huge data center projects around the country. But these Californians won’t get to vote on Nevada projects that could affect their electricity supply.
Corporate customers have reeled at the real costs of AI, which they discovered after Anthropic and other firms migrated from flat subscription fees to “per-token” fees, which charge by usage.
In a podcast in May, Uber Chief Operating Officer Andrew MacDonald said the company had not seen productivity gains commensurate with its spending on AI. And its chief technology officer said in April that Uber had spent its entire 2026 budget for Anthropic’s Claude platform by mid-March.
Expectations of the proliferation of AI across industries and in consumer applications were premised on a massive build-out of the data centers needed to support the technology. But most of the promised centers have run into snags. Local residents have turned out in force to resist the gargantuan installations — and their political leaders are listening.
Physical obstacles also have emerged. Local and regional electrical grids don’t have the capacity to fill the voracious power demands of big centers. Crucial electrical equipment is also in short supply. As Bloomberg reported, almost half of the U.S. data centers that were planned for this year are destined to be delayed or canceled. JPMorgan Chase has found that nearly two-thirds of the centers that were slated for completion next year haven’t yet broken ground.
Despite all this, the AI industry’s demand for capital has been unrelenting. Anthropic said on June 1 that it had made a confidential filing with the Securities and Exchange Commission for an IPO that may come before the end of the year. OpenAI is known to be contemplating a similar approach but hasn’t yet filed.
The sizes of the projected IPOs haven’t been made public, but they are likely to be based on the putative value of the companies in the private markets: Anthropic last month raised $65 billion from venture investors at a valuation of $965 billion, and OpenAI’s private market value is about $852 billion.
Neither company has yet reported a profitable quarter. Indeed, audited figures published last week by financial blogger Ed Zitron and the Financial Times indicate that OpenAI lost a staggering $38.5 billion in 2025 on revenue of $13.1 billion. (I asked OpenAI whether these figures were accurate but received no reply.)
Where the money will come from to cover these losses and power the industry ahead is an open question. But their appetite for capital appears to be insatiable; Monday, only days after raising $85 billion in its IPO, SpaceX launched a $20-billion bond issue.
That brings us to the general pitfalls of IPOs for ordinary investors.
Disney and OpenAI thought their billion-dollar deal would underscore the importance of AI for Hollywood’s future. Its ignominious collapse proves just the opposite
In the public mind, IPOs can be opportunities to get in on the ground floor of high-flying new ventures. The reality differs from the expectations, especially in the short- and middle term. The most recent IPO frenzy before now, the dot-com era of 1999-2000, tells the tale. In 1999, more than 480 companies went public, including about 289 internet companies. Three-quarters of the latter were unprofitable at the time of their IPOs.
That was the “highest concentration of money-losing IPOs in history,” by the reckoning of venture investor Trace Cohen, but that didn’t seem to stem the enthusiasm. About half of the 1999 IPO class went bankrupt by 2001, including one-time big names such as Webvan and EToys. An additional 20% were eventually acquired for less than the IPO price.
Those who view IPOs as stepping stones to great wealth, as it turns out, are making a category error. The IPO is typically a liquidation event for the benefit of founders, early investors and other insiders. They’re seizing the opportunity to cash out by selling their shares.
Accordingly, some of the smartest and most successful investors avoid buying into IPOs. Among them is Warren Buffett, the quintessential “value” investor.
Buffett explained to CNBC in 2019 why Berkshire Hathaway, his investment vehicle, had shunned the high-profile Uber IPO: “In 54 years, I don’t think Berkshire has ever bought a new issue,” he said. “The idea of saying the best place in the world I could put my money is something where all the selling incentives are there, commissions are higher, the animal spirits are rising, that that’s going to better than 1,000 other things I could buy where there is no similar enthusiasm … just doesn’t make any sense.”
Buffett was only applying a precept set down by his investment mentor, Benjamin Graham, in his classic primer “The Intelligent Investor.” As Graham wrote, “New issues have special salesmanship behind them, which calls therefore for a special degree of sales resistance.” Moreover, he wrote, “Most new issues are sold under ‘favorable market conditions’ — which means favorable for the seller and consequently less favorable for the buyers.”
At the time the book was first published (1949), the investment bank commission for IPOs was 7%, nearly twice the commission on sales of older shares. “Whenever Wall Street makes roughly twice as much for selling something new as it does for selling something old,” Graham observed, “the new will get the harder sell.”
The AI frenzy has been fueled by relentless hype, but the dud launched by OpenAI has its former enthusiasts wondering if they’ve been taken.
Things have changed since then. Commissions on existing share transactions have fallen to zero, and investment bank fees for the largest recent IPOs have fallen below 1%. The syndicate of investment banks underwriting the SpaceX IPO, which was the largest and most highly-touted IPO in history, accepted a fee of only 0.7%.
Conventional IPOs, in other words, have always been marketed at a time of the insiders’ choosing, and at a price favorable to them and backed by a turbocharged publicity campaign. The IPOs of OpenAI and Anthropic, like that of SpaceX, will also ride a mania for AI conjured out of vague and thus-far dubious promises of productivity gains from a largely untested new technology and hand-waving allusions to amazing possibilities.
This comes at the very moment when corporate users are discovering how difficult it is to justify their spending on AI and the building of AI infrastructure is running into public, physical and financial snags. So the question becomes not only where the sought-after capital will come from, but how it will be spent.
For these reasons, the AI IPOs have less a forward-looking than a fin de siècle aura about them. Rather than harbingers of a new world, they may be signs that insiders are hoping to get out while the getting is good — in other words, that the AI bubble is deflating in front of our eyes.
Insights
L.A. Times Insights delivers AI-generated analysis on Voices content to offer all points of view. Insights does not appear on any news articles.
Perspectives
The following AI-generated content is powered by Perplexity. The Los Angeles Times editorial staff does not create or edit the content.
Ideas expressed in the piece
The article contends that the planned IPOs of high-profile AI firms such as OpenAI and Anthropic look less like the dawn of a new era and more like a classic sign that insiders believe the AI bubble is near its peak and starting to deflate, with IPOs functioning as a way for early backers to cash out before sentiment turns.
It argues that the fundamental business case for AI has been overstated: many companies that aggressively adopted AI and laid off workers have found the promised cost savings illusory, in some cases rehiring staff in what critics label an “AI boomerang effect,” while corporate users such as Uber publicly report that AI spending has not produced commensurate productivity gains.
The piece stresses that the economics of AI usage are proving more punishing than advertised, as vendors shift from flat subscription models to usage-based “per-token” pricing, leaving enterprise customers facing unexpectedly high bills and making it harder to justify large-scale deployments.
Furthermore, the article maintains that the physical and political foundations for AI expansion are wobblier than the hype suggests: the expected surge in data centers has run into community opposition, power-grid constraints, and equipment shortages, with a significant share of planned U.S. facilities reportedly delayed or canceled, undermining the narrative of frictionless, exponential AI growth.
It highlights that despite sky-high private valuations for firms such as Anthropic and OpenAI, neither has posted a profitable quarter, and newly released figures indicating multibillion-dollar annual losses at OpenAI raise questions about how long investors will be willing to fund such cash burn and where future capital will come from.
The article links the current AI IPO wave to historical patterns from the dot-com boom, noting that late-1990s IPOs were dominated by unprofitable companies, many of which quickly went bankrupt or were acquired at losses to public investors, suggesting today’s AI flotations could similarly leave ordinary buyers holding the bag.
It emphasizes that, in practice, IPOs are typically liquidity events for insiders, not “ground floor” opportunities for small investors: the offering price and timing are chosen to benefit founders, early backers, and underwriters, echoing longstanding warnings from value investors such as Warren Buffett and Benjamin Graham about the heavy salesmanship and misaligned incentives around new issues.
While acknowledging that some mechanics of Wall Street have changed, such as lower underwriting fees and zero-commission trading, the article argues that the core dynamic remains: IPOs are pushed hardest precisely when enthusiasm is highest and skepticism is lowest, a pattern now visible in AI, where vague promises of transformative productivity gains are colliding with mounting real-world costs and constraints.
Taken together, the piece suggests that the imminent AI-related IPOs have a “fin de siècle” feel: rather than heralding a durable AI revolution, they may mark the moment when insiders try to exit amid fading returns, mounting losses, and growing doubts about whether AI can live up to its most extravagant economic claims, reinforcing broader commentary that the AI boom bears the classic hallmarks of a financial bubble.[1][2][3][5]
Different views on the topic
In contrast, some market commentators argue that AI is not primarily a speculative bubble but a long-term technological “wave,” contending that recent excitement reflects the early stages of a structural shift in computing—similar to past platform changes like the internet or smartphones—rather than a fleeting mania destined to collapse.[4]
Building on that view, opposing commentary suggests the most frothy activity so far is concentrated in infrastructure—such as chips, data centers, and cloud capacity—while the major AI application companies (including the very firms planning IPOs) are still in relatively early commercialization phases, implying that public listings may represent the start of a multi-decade buildout rather than the end of a craze.[4]
Additionally, this perspective holds that heavy capital spending and initial losses at leading AI developers should be seen as classic upfront investment in foundational technology, not as proof of a doomed model, pointing to historical cases where financially painful early years eventually led to dominant, highly profitable platforms once ecosystems and use cases matured.[4]
Some analysts in this camp also note that current concerns about constraints on power, hardware, and data-center buildout can be interpreted as signs of intense demand and rapid adoption, not of fundamental failure, arguing that such bottlenecks are common in breakthrough technologies and often spur innovation and infrastructure upgrades rather than precipitating collapse.[4]
Finally, these opposing views maintain that while AI-related stocks may well experience sharp corrections or sector rotations, the underlying advances in machine learning, automation, and software capabilities are likely to keep spreading across industries, so any bursting of a narrow valuation bubble in specific names would not negate the broader, long-run economic impact that AI proponents anticipate.[4]
