NVIDIA-Backed Firmus Grid Withdraws $5 Billion IPO After Investors Reject $30 Billion Valuation

Investors Reject a $30 Billion Valuation

Firmus Grid, the Australian AI data center developer backed by NVIDIA, withdrew its initial public offering on Friday, October 9, scrapping a New York listing that sought about $5 billion at a target valuation near $30 billion. The deal failed to draw sufficient demand from US institutional investors, who judged the price excessive for a company that reported $51 million in revenue for fiscal 2026 and had completed only 46 megawatts of the 912 megawatts of capacity in its build-out plans. The offering had been expected to rank among the largest in Australian corporate history. Firmus carried a $5.5 billion valuation in an April funding round that included NVIDIA and Coatue, then lifted it above $10.5 billion after raising roughly $2 billion more, before seeking more than $30 billion within a matter of months. The company applied a valuation framework comparable to the one US AI cloud provider CoreWeave uses, but built on expected operating profit two years forward. Some institutions indicated they would accept about $25 billion; none would agree to the final price. Underwriters including Bank of America, JPMorgan and Morgan Stanley cut the deal to roughly $3 billion and weighed a valuation range of $20 billion to $25 billion without success. Investors also objected to the absence of a lock-up for existing shareholders, which opened the door to heavy selling once the stock traded, and to the timing of a supply agreement with Meta for Southeast Asian compute capacity, announced just before marketing began alongside raised forecasts. Leonid Mironov, a portfolio manager at Gavekal Capital, criticized the transaction's structure, size, pricing and valuation method. The collapse hit related shares at once: Mas Group Holdings, a Firmus investor, fell as much as 30% intraday in Sydney. Firmus operates two leased data centers, in Melbourne and Singapore, and had planned five more across the Asia-Pacific region. It is now weighing a smaller private raise from existing investors and other international listing venues; its shareholder letter cited market volatility and conditions that did not fairly reflect long-term growth prospects, and partner CDC confirmed that a joint 1.6 gigawatt AI data center plan is no longer proceeding.

Lumentum's Order Book Runs Into 2029

Lumentum has sold out its manufacturing capacity for AI data center optical components into early 2029, chief executive Michael Hurlston said, in a disclosure carried by WuBlockchain's daily brief. The US company builds the high-speed optical devices that move data between chips and racks inside AI clusters, and demand now outruns what its factories can deliver for years: for some products, roughly 70% of orders cannot be fulfilled through 2027, while for others about 30% of demand remains uncovered through 2028. The company plans to invest at least $350 million in key plants in the Tokyo metropolitan area and adjacent facilities to raise output, and has already widened capacity at its United Kingdom site. Hurlston cautioned that new capacity typically takes three to five years to build. Several large cloud providers have agreed to absorb the capital-expenditure risk of the expansion, an arrangement that reserves future supply for the customers while keeping the factory buildout off Lumentum's own financing plan. The booked orders extend revenue visibility deep into the decade, even as the long construction cycle caps how quickly unmet demand converts into booked sales. The shortfall percentages also put numbers on a bottleneck operators feel directly: optical transceivers are among the components whose supply lags GPU deployments across the industry.

Contracted Revenue Funds, Projected Revenue Does Not

Friday's two stories mark the boundary of what AI capital markets will currently finance. Lumentum needs no raise: its capacity is contracted years ahead and its cloud customers are pre-funding the factories. Firmus, with $51 million of revenue behind a $30 billion ask, could not clear even a reduced $20 billion to $25 billion range, and Mas Group Holdings' 30% slide shows where the exposure surfaces for listed holders. The same cycle carried OpenAI telling investors its annualized revenue ran near $50 billion at the end of September, with at least $70 billion projected by the end of 2026, growth of roughly 40% that rests on enterprise expansion. That projection, not a booked order book, is what private capital is now being asked to underwrite.