- Firmus Grid, the Nvidia-backed cloud-computing company, abandoned plans for Australia’s largest IPO in three decades after investors refused to pay its asking price. The company had sought a $30.4 billion valuation for a $5 billion raise, but bookbuilding collapsed when institutional investors did the math: Firmus operates just two data centers, with five more in development. By contrast, CoreWeave, a direct competitor already listed on Nasdaq, operates 51 facilities and trades at $37 billion. That means Firmus was asking for 82% of CoreWeave’s value while running roughly 4% of the capacity. The math didn’t work. Firmus said it would instead pursue further private investment, a retreat signaling that even the hottest sector can’t escape valuation discipline.
- Bond yields becoming an invisible headwind. With global yields surging in recent months, institutional investors are demanding higher compensation for risk—and that price discipline is flowing into equity valuations too. When bonds pay more, stocks have to offer better returns to justify the same level of risk. For data center operators with uncertain profitability, that squeeze is real. CoreWeave’s stock fell nearly 8% Thursday amid broader worries about whether AI infrastructure companies can actually turn their massive capex into revenue. Higher debt costs also matter: Firmus was planning to finance data center buildouts with debt, and rising bond yields make that financing more expensive.
- The collapse happened fast. Just two months ago, institutional investors Blackstone and Coatue had valued Firmus at $10.5 billion when they led a $2 billion funding round. Then the IPO price target jumped to $30.4 billion—nearly 3x in eight weeks. That disconnect between private and public valuation is the real story. Blackstone and Coatue saw the numbers and probably understood the economics; institutional IPO buyers saw a company with minimal revenue history and massive capex requirements and balked. The gap between insider and outsider valuations can persist in venture capital, but public markets have a way of forcing reality.
- Firmus still has a story: demand for AI compute outstrips supply, Nvidia is expanding capacity globally, and companies like OpenAI and Meta are willing to pay for computing power. The retreat doesn’t invalidate that narrative. But it does show that investors want proof of business model, not just proof of demand. With only two operational facilities, Firmus can’t show unit economics at scale yet. CoreWeave, with 51 operating facilities, can. That gap between potential and proven is worth billions in valuation, and the IPO market just enforced it.
What Happened?
Firmus Grid pulled plans for a $5 billion initial public offering after institutional investors rejected its proposed $30.4 billion valuation. The Australian cloud-computing company, backed by Nvidia, had targeted Australia’s largest IPO since the Telstra partial privatization in 1997. The bookbuild collapsed when investors compared Firmus’s two operational data centers to CoreWeave, a listed competitor operating 51 facilities with a $37 billion market cap. Firmus countered by pointing to its further five data centers in development across Australia, Malaysia, and Indonesia, as well as strong customer relationships with OpenAI and Meta Platforms. But the path from two to seven facilities does not yet prove the business model works at scale. The company said it would instead pursue further private funding. The retreat comes ahead of Anthropic’s planned blockbuster IPO and amid broader investor caution about AI infrastructure valuations as bond yields have risen and questions emerge about the profitability of data center operators.
Why It Matters?
Valuation gaps between private and public markets reveal different information sets and risk tolerances. Venture capital investors place bets on future potential; public market investors demand near-term path to profitability. Firmus’s collapse from $10.5 billion (private valuation two months ago) to withdrawn IPO (proposed public valuation $30.4 billion) shows the gap widened rather than closed—a red flag. CoreWeave trades at $37 billion with 51 facilities; Firmus was asking for $30 billion with 2 facilities and a development pipeline. The math doesn’t work until Firmus proves it can execute and monetize at scale. The broader message: AI infrastructure demand is real, but investors are no longer willing to pay for just the narrative. Higher bond yields have made equity investors more disciplined about returns on capital. Rising debt costs also threaten data center economics—these businesses require massive upfront capex financed by bonds and equity. If yields stay high, fewer projects pencil out economically. CoreWeave’s 8% drop Thursday reflects investor worry about whether the entire neocloud cohort can earn adequate returns on the capital being deployed.
What’s Next?
Watch Anthropic’s IPO timing. It was planned for pre-Thanksgiving November, but if investor appetite for AI infrastructure is cooling, Anthropic might adjust its valuation expectations or timeline. If Anthropic prices conservatively and pops on opening, that signals the market wanted deals, not hype. If it prices conservatively and trades flat or down, that confirms investor caution about AI valuations broadly. Track CoreWeave for signs of operational stress. If its stock stabilizes and facilities ramp to revenue, that validates the neocloud model and opens the door for Firmus to try IPO again at lower valuation. If CoreWeave continues sliding, it suggests the market is pricing in execution risk across the sector. Monitor bond yields: if they fall, financing becomes cheaper and Firmus’s math improves. If they stay elevated or rise further, the IPO window stays closed. Finally, watch for consolidation. If Firmus can’t go public and can’t raise more private capital at acceptable terms, Nvidia might acquire it (Nvidia is already a customer, hardware supplier, and shareholder) to integrate it vertically, or a larger infrastructure play might buy it for the pipeline.
Affected Tickers and Coins: NVDA | CWEQ
Source: Wall Street Journal
