Firmus Reportedly Raises $505M — And Its $5.5B Valuation Is Already Stale
April 6, 2026 — Firmus Technologies announced today it has raised $505 million in a round led by Coatue Management, with Nvidia also participating, valuing the Australian AI data center builder at $5.5 billion, as reported by Bloomberg. The company has now raised $1.35 billion in six months. The cash funds rapid deployment of AI hardware across Australia and Singapore under Nvidia's Vera Rubin DSX reference architecture — Nvidia's next-generation chip platform shipping in the second half of this year.
The headline valuation, however, is already behind reality. Firmus secured a US$10 billion debt facility in February 2026 led by Blackstone, pushing its implied value well above $5.5 billion. In March it announced a multi-billion-dollar, multi-year contract for ~18,400 Nvidia GB300 GPUs at its Melbourne facility. A Tasmania campus targeting 36,800 GPUs is under construction. An ASX IPO is expected mid-2026.
Strategically, Firmus is the clearest southern-hemisphere expression of Nvidia CEO Jensen Huang's sovereign AI thesis — local data centers keeping national data within national borders, powered by renewable energy. Coatue, which manages over $70 billion and has backed OpenAI and Anthropic, lends institutional credibility. Blackstone, the world's largest alternative asset manager, provides the debt underpin.
Yet investors should separate the story from the underwrite. The hyperscale customer signed in March remains unnamed. Pricing mechanics, power cost passthroughs, and cancellation rights are undisclosed. The rumored IPO range of A$8–12 billion would be a dramatic markup from even the corrected valuation — before a single quarter of public financials. A note of caution: Nvidia backing companies that also buy its chips is a dynamic some investors have flagged as circular, something Nvidia has pushed back on. The US$10 billion debt is not automatically bullish for equity — project-finance debt means grid delays, transformer bottlenecks, or GPU timing slippage hurt equity holders first.
CoreWeave: The Most Real Story, With the Narrowest Path
CoreWeave (CRWV, $80.94) reported 2025 revenue of $5.131 billion, adjusted EBITDA of $3.093 billion, and a $66.8 billion revenue backlog. Microsoft accounted for 67% of 2025 revenue. OpenAI has committed up to $11.9 billion through October 2030.
The pivotal recent development was not ARK Invest's purchase of 41,830 shares on March 30 (~$2.89 million) — that is sentiment. It was CoreWeave closing an $8.5 billion delayed-draw term loan rated A3/A(low): the first investment-grade, non-recourse facility of its kind for HPC infrastructure, tied directly to contracted customer demand. That is infrastructure capital markets creeping into compute — structurally lowering cost of capital and converting customer contracts into something closer to collateralizable cash flows. Nvidia reinforced alignment by investing $2 billion at $87.20/share in January 2026.
The bull case is real. So is the bear case: 67% concentration in one customer is not diversified utility economics. At ~$68 billion market cap (roughly 13x 2025 revenue), the equity only works if financing innovation continuously outpaces compute commoditization. That is a narrow corridor, not a guarantee.
Intel EMIB: The Most Underappreciated Second-Order Trade
TSMC's CoWoS packaging — the process physically connecting memory and compute on flagship AI chips — remains the dominant bottleneck in the AI supply chain. Intel's EMIB is the most credible alternative: analyst estimates put power efficiency at ~0.30 pJ/bit vs. CoWoS's ~0.56 pJ/bit, wafer utilization at 90% vs. ~60%, and packaging cost in the "few hundred dollars" range versus $900–1,000 for CoWoS on a Rubin-class processor. Intel targets package areas exceeding 12x reticle size by 2028, against CoWoS-S's current ~3.3x.
Hyperscalers are not seeking alternatives because TSMC is technically inferior. They want supply-chain resilience, US-based capacity, and relief from CoWoS pricing power and Taiwan geopolitical risk. EMIB just needs to be available, scalable, and good enough — criteria it increasingly meets. At $50.78 (~$155 billion market cap), Intel is still priced as a CPU-centric turnaround. The packaging optionality is not in the stock.
The Investment Hierarchy
One company is priced off public evidence and financing innovation. One is marketed off private scarcity and forward promises. One is monetizing the market's desperation for a second packaging source.
Ranked by evidence quality: CoreWeave, Intel packaging, Firmus. Ranked by sentiment overshoot risk: Firmus, CoreWeave stock, Intel. Ranked by market underappreciation: Intel packaging, CoreWeave's financing model, Firmus as a sovereign-compute utility.
CoreWeave is riskier than bulls admit but more investable than bears claim — the disclosures are real. Firmus today looks explosive; but today's $5.5 billion headline is already stale, the IPO range implies further aggressive step-ups, and public investors would be paying private-market multiples without public-market disclosure. Intel packaging is the most overlooked: much of TSMC's margin structure is scarcity rent, and scarcity invites alternatives.
The real 2026 AI trade is not compute scarcity in the abstract. It is who can convert compute demand into bankable assets without being crushed by power costs, packaging constraints, customer concentration, and financing structure. Only one of these three has demonstrated that publicly. The other two are still asking investors to trust the story.
not investment advice