An Unlikely IPO Collapse Sends a Warning to Nvidia and Micron Investors

An Unlikely IPO Collapse Sends a Warning to Nvidia and Micron Investors

The AI boom just got a warning shot from a place nobody was watching. An Australian data center operator pulled the plug on what would have been one of the world's largest IPOs this year — and the bad news arrived alongside soft revenue signals from OpenAI and a wave of bearish bets in the chip options market.

A $5 billion IPO pulled at the last minute

Firmus, an Nvidia-backed AI data center operator based in Australia, withdrew its planned listing on the Australian Securities Exchange on October 9, citing "market volatility and conditions." According to Reuters, the IPO would have raised about US$5 billion at a valuation of roughly US$30.6 billion — the second-largest new share sale in Australian history and the fourth-largest IPO globally this year, behind only SpaceX, CXMT, and Cerebras. Investors began pulling orders on Wednesday, and the company pivoted to raising money in private markets, with a potential Nasdaq listing to follow, according to a person involved in the deal. The company's official line is that it will "pursue capital from the private markets and consider alternative public and private market options."

Why the demand didn't hold

The deal's pricing tells the story. Firmus raised US$2 billion in August at a US$10.5 billion valuation, backed by Nvidia, Coatue Management, Blackstone, and Jane Street. The IPO asked public investors to pay nearly three times that valuation less than three months later, with shares priced at A$11 — reports later mentioned demand being tested as low as A$5.50. For investors already scrutinizing the sustainability of AI infrastructure spending, the math didn't work.

Demand-side warning signs are multiplying

AI labs are passing hardware costs on to customers, and customers are pushing back. OpenAI recently introduced a US$500-per-month plan for uncapped use of its flagship models — a 150% increase over its previous top tier of US$200. A Financial Times report last week suggested OpenAI's annualized revenue is actually tracking at around US$50 billion, below the US$70 billion figure previously reported elsewhere — possibly a sign customers are pulling back after price hikes. Uber, Walmart, AT&T, and even Amazon have placed restrictions on employee AI use this year to prevent budget blowouts. And a UBS survey found that 60% of businesses are diverting some AI workloads to cheaper, more efficient models, which typically consume less computing power — and fewer chips.

A crowded supply side for Micron

For Micron, the pressure is coming from both directions. Nvidia CEO Jensen Huang confirmed in June that all three major memory makers — Samsung Electronics, SK Hynix, and Micron — had been approved to supply HBM4 high-bandwidth memory for Nvidia's next-generation Vera Rubin platform. That's good in that Micron stays in the game, and bad in that Nvidia now has three suppliers to play against each other. Meanwhile, Counterpoint Research data shows Micron's share of the stacked-memory market slipped to 18% in the second quarter, while Samsung doubled to 33%. Micron plans to double its monthly output of stacked memory to roughly 100,000 wafers by year-end — adding capacity while losing share.

The options market is already voting

Traders saw it first. On October 9, more than 180,000 put contracts traded on the SMH semiconductor ETF by midday — versus only 50,000 calls — with US$46 million in premium tied to puts against US$26 million in calls, according to ThinkOrSwim and SpotGamma data cited by CNBC. The put-to-call open interest ratio rose to 1.95, its highest since early August. One Nvidia trade stood out: 100,000 put contracts with a US$180 strike, expiring January 2027, for US$21 million — the largest single trade in the stock's options market that day. On the Micron side, roughly US$270 million in premiums was associated with put purchases, including a series of deep in-the-money puts expiring June 2028.

What comes next

Nvidia says the top-five hyperscalers are on track to spend about US$800 billion on AI data center infrastructure in 2026 and could spend another US$1.3 trillion in 2027. In theory, the buildout continues as long as the cloud giants believe demand is real. Over the next month, those same companies will report results for the quarter ended September 30 — and if capital expenditure guidance pulls back, a semiconductor demand slowdown is almost certainly on the horizon. Valuations have already adjusted: Nvidia trades at a price-to-earnings ratio of just 29.1, less than half its 10-year average of 61.3, while Micron's P/E of 13.9 sits well below the S&P 500's 23.3 and the Nasdaq-100's 34.6.