評価額だけ先に開通させるな|Firmus IPO撤回とAIバブルの身分証検査|Firmus’s Withdrawn IPO: Valuation Cannot Outrun Infrastructure

The two-day collapse Australian AI-infrastructure company Firmus Technologies, backed by NVIDIA, Blackstone, Coatue and Jane Street, raised US\$2 billion at a valuat

Share

English · 日本語

アーカイブ移行日:2026年11月11日。原記事の初出日は未確認。 / Archived 11 Nov 2026; original publication date unconfirmed.

The two-day collapse

Australian AI-infrastructure company Firmus Technologies, backed by NVIDIA, Blackstone, Coatue and Jane Street, raised US\$2 billion at a valuation above US\$10.5 billion in August 2026. Two months later it sought an equity value of roughly A\$44 billion, or US\$31 billion.

Weak demand forced a roughly 25% offer-price cut and reduced proceeds from about A\$7.9 billion to A\$5.9 billion. Investors still refused the deal, and the IPO was withdrawn.

Why investors said no

Firmus had completed only about 42MW while promoting a pipeline approaching 1GW. Contracts have value, but land, power, cooling, GPUs, permits, financing, construction and customer demand must all arrive before capacity becomes productive.

The company was also changing strategy toward self-development in Southeast Asia. Public investors were asked to pay a completed-business valuation for a plan still under construction, at nearly three times the valuation of two months earlier.

The strange metric

The valuation used EV+1/EBIT+2: enterprise value based on estimated net debt one year ahead divided by projected operating profit two years ahead. The logic reflects a construction cycle, but the risks are asymmetric. Debt often arrives on schedule; profit may not. If power, construction or GPU deliveries slip, EV+1 becomes real while EBIT+2 moves farther into the future.

Fear defeated FOMO

Famous sponsors, major customers and a near-1GW pipeline once triggered fear of missing out. This time, investors feared being left holding an overpriced dream. The market is moving from contracted to completed capacity, EBITDA to EBIT and free cash flow, GPU counts to utilization, and famous backers to financing durability.

The blunt take

Investors did not reject AI. They rejected the price tag on an unfinished power-and-server complex. A good business is not a good investment at every price, and new capital for construction must be separated from liquidity for early shareholders.

Is the AI bubble bursting?

This is an identity check inside the bubble, not proof that AI demand has vanished. Cash-rich hyperscalers differ from companies dependent on repeated IPOs, bonds or GPU-backed loans. But broader neocloud financing stress could delay orders for GPUs, power, cooling and optical connectivity.

Japan market

Sentiment can hit SoftBank Group, Advantest, Tokyo Electron, Disco, Furukawa Electric, Fujikura, Sumitomo Electric and data-center, power, cooling and real-estate names. One failed IPO is not enough to cut earnings forecasts; watch financing terms, project delays, utilization and cancellations.

Furukawa-Oyaji ruling

“Connect the optical fiber first. Do not connect the valuation before the facility.”

Verdict

Firmus did not burst the AI bubble. It pressed the balloon and showed that the skin was thinner than expected.


Original Notion archive