business 6 min read

Anthropic IPO Shows the Cracks in AI Euphoria

Seoul's hesitation and Beijing's market correction both tell the same story: AI companies have raised enough to build the future, but perhaps not enough to sustain it. The real question is who profits and who gets squeezed when these giants start answering to public markets.

  • Artificial Intelligence
  • Korea Stocks
  • IPO Market

The Paradox Behind Anthropic’s Big Listing

Anthropic is preparing for one of the largest technology debuts of the decade — possibly larger than OpenAI’s eventual exit. The numbers are staggering: roughly $90 billion in anticipated proceeds, hundreds of billions in multi-year computing contracts with Amazon Web Services and Google Cloud extending through 2045, and a customer base that includes Microsoft, Salesforce and countless startups building on Claude. Yet beneath the surface of these figures lies a question that Korean and Chinese investors seem to have already answered: how much is enough?

From Seoul to Beijing, the early signals were clear. YJ Investment & Securities warned its clients ahead of the November 19 listing that large-scale IPOs historically mark late-cycle exhaustion, pointing to the dot-com bubble (2000), Google (2004), Alibaba (2014) and the SPAC mania (2021) as precedents. In each case, the subsequent year brought either declining S&P 500 returns or a dramatic slowdown from single-digit to sub-single-digit gains. The same pattern held in China, where newly listed artificial intelligence companies saw their shares collapse by as much as 40 percent within months of going public.

What separates this cycle from earlier ones is not the size of the raises — though the $240 billion in projected US IPOs this year does exceed the $106 billion during the 2000 peak — but the composition of the demand behind them. Anthropic alone has locked in computing purchases valued at $895.4 billion, according to its Nasdaq filing. Its current cash position sits at $131.7 billion, meaning even after the listing and assuming a generous $90 billion raise, it will still face a shortfall of roughly $670 billion over the coming decades. That is not a problem solvable with another equity round. It requires either extraordinary revenue generation, deep strategic partnerships that effectively subsidize the gap, or the kind of acquisition-driven consolidation that has not yet become visible in the sector.

Who Wins, Who Loses, and Who Gets Left Behind

The winners in this equation are easy to identify: the hyperscalers supplying compute, the infrastructure partners financing construction, and the venture firms sitting on deeply discounted stakes who will soon have a public market to exit into. AWS and Google Cloud each hold multi-billion-dollar contracts with Anthropic that lock in decades of predictable revenue. For those companies, the listing is a net positive regardless of how the stock trades on day one.

The losers are less obvious but equally consequential. Retail investors arriving after the initial pop will inherit the burden of dilution risk — the probability that Anthropic must issue additional shares to fund operations, further compressing per-share value. Institutional investors who waited for the offering, hoping to buy at a discount to private rounds, may find themselves bidding against the very companies whose products depend on Claude’s continued dominance. If Anthropic succeeds as a platform, competitors like OpenAI, Mistral and xAI suffer. If it struggles to convert contracts into profit, the entire thesis for AI infrastructure spending comes into question.

Chinese and Hong Kong markets already absorbed this lesson in abbreviated form. Over two dozen AI-focused listings in 2024 and 2025 saw average post-issue corrections exceeding 30 percent, driven by the same dynamic: companies raising far more than they could productively deploy, then facing margin pressure as capital expenditures outpaced subscription revenue. The Hang Seng Technology Index has underperformed the broader market by roughly 18 percent over the past twelve months, a gap driven disproportionately by unprofitable growth names rather than mature tech firms.

South Korea occupies a middle ground. Domestic institutions showed notably muted appetite for pre-marketing allocations tied to US-based AI names, with several pension funds explicitly citing valuation dislocation as a reason to stand aside. Meanwhile, Korean construction and materials companies traded higher on speculation that US data-center regulation would slow deployment timelines and reduce demand — a trade that assumes the boom is ending sooner rather than later.

What the Numbers Actually Say

The empirical record is sobering. Between 2000 and 2021, every major cluster of large-cap technology listings was followed by a year in which the S&P 500 declined or posted returns below 10 percent. The magnitude varied: the 2000 post-bubble decline reached 14.1 percent in the first year and 17.3 percent the second. The 2021 follow-through was milder at 9.7 percent but still unmistakable. In periods where listings surged without corresponding earnings growth — the defining feature of the SPAC era — underperformance was steeper and longer-lasting.

Anthropic’s profile differs from those earlier episodes in at least one critical respect: its revenue trajectory. Unlike dot-com-era companies selling advertising clicks or SPAC shells with no underlying cash flow, Anthropic reported approximately $3.5 billion in annual recurring revenue in 2025, representing year-over-year growth exceeding 150 percent. The gross margin stands near 70 percent, healthy for a software-intensive business but thin compared with legacy SaaS platforms operating at 80 percent or above. More importantly, the path from $3.5 billion to the tens of billions required to service $895 billion in infrastructure commitments is dramatically steep.

Even generous projections suggest Anthropic would need to grow revenue at a compound annual rate of roughly 40 percent over the next decade — faster than Amazon, Microsoft or Alphabet achieved during their most aggressive expansion phases. Revenue multiples imply a post-listing market capitalization north of $1 trillion, a figure that assumes perfect execution and zero competitive encroachment.

The Supply-Side Story No One Is Telling

The more interesting question is not whether Anthropic can justify its valuation, but whether the entire ecosystem can sustain the scale of capital now being deployed. Global hyperscaler capital expenditure across Amazon, Microsoft, Google, Meta and Apple is tracked at approximately $280 billion annually and rising. Add in Chinese spending by Baidu, Alibaba and Tencent, plus European public-private initiatives, and total infrastructure outlays approach $400 billion per year.

Semiconductor capacity constraints remain the binding bottleneck. TSMC’s advanced-node utilization hovers near 95 percent, with delivery queues extending 18 to 24 months for the most specialized GPU configurations. Each new AI training cluster consumes roughly 50 megawatts of power — equivalent to the electricity usage of a mid-sized city — and the permitting timeline for new data-center sites in Virginia, Oregon and Taiwan routinely exceeds three years.

These are not problems that resolve themselves through additional equity fundraising. They require either breakthrough advances in chip efficiency, geopolitical shifts that unlock new manufacturing capacity, or a rationalization of competing infrastructure builds that leaves fewer redundant deployments. None of those scenarios appears imminent.

What Happens Next

Anthropic’s listing will proceed. Regulatory approval has been secured, roadshow momentum is strong and the pricing window opens in mid-November. The more likely flashpoints are downstream: earnings calls revealing slower-than-expected contract conversions, follow-on offerings priced at discounts to the IPO, or competitive pressure from OpenAI’s anticipated entry into public markets later in 2027.

For investors, the practical takeaway is simple. Companies that can convert massive capital raises into durable free-cash-flow growth tend to reward patience. Those that cannot become cautionary tales written in diluted share counts and shrinking margins. The current crop of AI superstars occupies the former category only if they execute with unusual precision over an extended horizon — a condition supported by historical precedent, not by wishful thinking.

Seoul and Beijing already voted with their portfolios. The rest of the world is just catching up.