The AI Buildout Runs on a Circle. The Anthropic IPO Is the Attempt to Break It.

In late July, SK Hynix reported the largest quarterly profit in its history. The same week, its stock crashed 20%, dragging South Korea's KOSPI index into a bear market ^[1][2]. The company that makes the high-bandwidth memory chips every AI data center runs on had never been more profitable — and the market had never been more skeptical of what those profits were worth. Jim Cramer put the contradiction plainly.

SK is more about margin calls and gambling than it is about DRAM pricing and capacity additions — Jim Cramer

He was describing a single stock's volatility, but the diagnosis reaches further. What the KOSPI crash was pricing was not a collapse in AI demand. It was a question about whether the financial structure that funds the buildout can hold together long enough for revenue to arrive. SK Hynix's own management made the condition explicit. The company is in the middle of a $720 billion expansion to build the world's largest network of memory factories, including a $500 billion supply deal with Nvidia ^[3]. But its executives told investors the investment's durability beyond next year depends on one condition.

As these investments are supported by revenue generated from AI services, the momentum in memory demand is expected to persist. — SK Hynix

That revenue has not yet materialized at the scale required. And SK Hynix is already adjusting. The company said it is slowing its transition to HBM4, the next generation of high-bandwidth memory, because the production forecasts for Nvidia's Rubin chip — the GPU that will use HBM4 — are softening.

Since production forecasts for Nvidia's next-generation chip 'Rubin,' which will be equipped with HBM4, are also trending downward, there is no reason to accelerate the transition to HBM. — SK Hynix

The memory supplier at the very bottom of the AI hardware chain is telling the market that demand at the frontier is not accelerating the way the buildout plans assumed. This is not a demand-cratering signal. Cloud revenue is growing fast — Google Cloud up 82% year over year, AWS up 37% — and the physical capacity deficit is real: only 8.9 gigawatts of data center capacity came online in 2025 against 21.1 gigawatts of demand ^[4][5]. The question is not whether anyone wants the compute. It is whether the money to build it can keep flowing. One layer up, the debt numbers are starting to strain the balance sheets that carry them. OpenAI's cloud partners — the companies building the data centers the startup rents — have accumulated roughly $100 billion in debt to fund that infrastructure. OpenAI itself expects around $20 billion in revenue this year ^[6]. The ratio is five to one, and it is not the most extreme figure in the stack. Oracle has issued $18 billion in bonds with maturities as long as 40 years to fund data centers for OpenAI, pushing its total debt to $91 billion. Interest payments now consume more than 20% of its operating income ^[6]. Alphabet, which raised its 2026 capital spending guidance to $205 billion, posted its first-ever negative free cash flow — minus $5.9 billion — in the same quarter. Meta's free cash flow plunged 91% to $784 million ^[7]. These are not startups burning venture capital. These are some of the most cash-rich companies in the world, and the physical buildout is now consuming cash faster than their businesses generate it. Apollo's chief economist, Torsten Slok, estimates the total AI infrastructure buildout may require more than $2 trillion in debt — but traditional bond markets can only absorb about $1 trillion through 2030, leaving a trillion-dollar gap that would need to be filled by private credit ^[8]. The reason the debt keeps piling up is that the companies lending it are also the companies that stand to gain from it. The AI buildout runs on a circle. Nvidia reported $13 billion in investment gains from its stakes in OpenAI, Anthropic, and other AI infrastructure companies — the same companies that buy its chips ^[9]. Amazon derived 65% of its net quarterly income from its Anthropic stake. Across the Magnificent Seven, $134.6 billion — 42% of $315.6 billion in second-quarter profit — came from investment gains, not operations ^[9]. The structure is self-reinforcing until it is not. Nvidia invests in Anthropic. Anthropic uses the money to buy Nvidia chips and rent cloud capacity from Amazon and Google. The chip purchases generate Nvidia's revenue; the cloud contracts generate Amazon's and Google's. The rising valuations of the labs generate investment gains for the same companies. Everyone's profits depend on everyone else's spending, and the spending depends on the conviction that the revenue will eventually justify it. The counter-evidence is not trivial. Revenue from products built on Google's AI models grew nearly 800% year over year, Sundar Pichai told investors ^[10]. Amazon CEO Andy Jassy says AWS equipment purchases break even in under three years ^[4]. UBS analyst Rickie Fowler calls the Nasdaq "outright cheap," pointing to a $2.4 trillion cloud revenue backlog against $650 billion in capital spending ^[11]. The demand is genuine. The question the KOSPI crash raised is whether the financing model can bridge the distance between the spending and the revenue. This is what the Anthropic IPO is for. The company is targeting a valuation of $2 trillion to $3 trillion when it lists in October ^[12]. Its valuation has risen from $61.5 billion in March 2025 to $183 billion in September, $380 billion in February 2026, $965 billion in May, and now the $2-to-3 trillion target — a 33-to-49-fold increase in 17 months, while revenue grew roughly ninefold over the same period ^{[13][14][15][12]}. Valuation is outpacing revenue by a factor of four. The IPO is not merely a liquidity event for early investors. It is the attempt to break the circle. The private financing loop — Nvidia to Anthropic to Nvidia, Amazon to Anthropic to Amazon — can only scale so far. Public markets represent a new pool of capital, one that does not depend on the same circular logic. Goldman Sachs CEO David Solomon, whose bank is positioned to underwrite the wave, described the market mood as one where greed is running ahead of fear ^[16]. The bet is that public investors will price Anthropic at a level that validates the entire stack beneath it. Anthropic's own CEO has already described what happens if they do not.

If my revenue is not $1 trillion, if it's even $800 billion, there's no force on Earth, there's no hedge on Earth that could stop me from going bankrupt if I buy that much compute. — Dario Amodei