Nvidia-backed Firmus pulls its $30 billion IPO after demand collapses

The biggest AI-infrastructure listing of the year didn't get delayed — it got pulled entirely, after two price cuts in two days found no takers.

Nvidia-backed data-center operator Firmus withdrew its Australian IPO on Friday, abandoning what would have been the country's second-largest listing ever after slashing the offer price twice in 48 hours. The company had been marketing shares at A$11 each to raise about $5 billion at a valuation near $30 billion — roughly three times the $10.5 billion mark it carried just weeks ago, when Nvidia, Coatue and Blackstone put $2 billion into it. On Wednesday bankers cut the price to A$9 and shrank the raise; on Thursday they cut again to A$5.50; by Friday the offer was dead. Australia's big super funds, the natural buyers for a deal this size, refused to bite, and Firmus said proceeding "was not in the best interests of the company and its shareholders," sending it back to private markets instead.

The whole saga ran about five days from bookbuild hype to withdrawal, but the collapse itself was brutal and fast — the two cuts and the pullback all landed inside roughly 48 hours, after the company had publicly touted strong subscription interest. The pricing was always the problem: at the original offer, Firmus was valued at around 600 times its fiscal 2026 sales, with roughly $30 billion of debt propping up an enterprise value near $60 billion, according to analysts cited in the Japanese press. One fund manager's verdict to the Guardian was blunt: "If it got on to the ASX it was going to end in tears."

Why it matters: Firmus never traded, so there's no stock crash to point at — but this is the first hard market judgment on whether public investors will fund AI data-center buildouts at the valuations private investors have been paying. A valuation that tripled in seven weeks now has a public-market price ceiling, and the timing is awkward: a wave of AI-infrastructure listings is lining up behind it for a template. Investors have already balked at one mega-listing this quarter — SoftBank's SB Energy delays a $50B IPO as investors balk — but at least that one was postponed rather than abandoned. The $5 billion that didn't show up has to come from somewhere, and private markets — where Nvidia and friends are comfortable writing checks — is the obvious stopgap.

What to watch: whether Firmus actually raises privately at anything close to the failed IPO valuation, and whether the next AI-infrastructure prospectus opens at a lower number.

Is the public-market funding window for AI data centers closed for this cycle — or was Firmus just mispriced? Tell us in the comments.

Two fixed opening tokens push a base model past its RL version

A new paper from MIT, UC Berkeley, Washington University and the Allen Institute for AI says much of what reinforcement learning teaches a model may come down to how it starts an answer. One good prefix, fixed in advance, is enough to close the gap. Fix the first two tokens and Olmo-3-7B beats its own RL-trained twin. The paper — "Base Models Can Reason By Taking a Cue From Training Data" — tested what happens when researchers pre-fill a model's response opening instead of letting it choose. O

Senate report: hyperscalers misled the public on data center costs

A yearlong Senate investigation just put Congress's own numbers under the claims AI data centers sell to towns and ratepayers. Also: the Times finally counts what Anthropic's agents submitted to the State Department, and StepFun sets an open-weights date for its top-ranked flagship. A yearlong Senate investigation led by Senators Warren, Van Hollen, and Blumenthal concludes that some of the biggest hyperscalers misled the public about what their AI data centers cost everyone else. Staff querie

AI inference is redrawing the storage hierarchy

The money has followed GPUs for years, but at GMIF 2026 in Shenzhen the argument was about everything behind them — and the numbers backing it are getting hard to ignore. China's 140 trillion daily token calls are forcing a redesign of the storage stack. Now in its fifth year, the Global Memory Innovation Forum gathered storage makers, analysts and chip designers around one theme: inference, not training, is now the workload that dictates hardware. China's daily token calls had already passed

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%

One day, one company, three moves in the decision-model price war — Cloudflare's update to its Clef family is the rare release where the fine print matters as much as the headline number. Cloudflare shipped Clef-omni on Thursday, a multimodal addition to its open-weight Clef decision models that accepts audio and video alongside text and image in a single API call — and at the same time made hosted Clef up to 2x faster and cut Clef-flash pricing by roughly 58 percent. Clef-omni takes WAV or MP

Read more

Two fixed opening tokens push a base model past its RL version

Senate report: hyperscalers misled the public on data center costs

AI inference is redrawing the storage hierarchy

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%