The AI Reality Check: What APAC Founders Can Learn from Firmus Grid’s Aborted $5 Bn IPO

In a dramatic pivot that sent shockwaves through the Asia-Pacific tech ecosystem, Nvidia-backed AI infrastructure provider Firmus Grid officially cancelled its planned $5 Bn initial public offering (IPO).

The corporate vehicle, which operates as Firmus Technologies under co-founders Oliver Curtis and Tim Rosenfield, formally withdrew its listing application from the Australian Securities Exchange (ASX). Bookbuilding hit a steep wall of global investor scepticism when fund managers baulked at an implied equity valuation of $30.6 Bn.

This targeted market cap was nearly triple the $10.5 Bn-plus valuation achieved in an August private round, where the company raised $2 Bn from a heavyweight roster of backers, including Nvidia, Coatue Management, Blackstone-managed funds and Jane Street.

For founders, tech builders and macroeconomic strategists across APAC, the sensational implosion of what would have been Australia’s second-largest corporate float offers a profound lesson. It clearly separates two ideas often treated as interchangeable during the AI infrastructure boom: the reality of compute demand versus the capitalise-it-today mindset of early-stage private market multiples.

The IPO was not withdrawn due to a lack of demand for AI computing capacity. Instead, public equity markets sharply differentiated between future contractual promises and current earnings.

That is not surprising. Although private capital is designed to absorb early-stage, long-duration execution risks, public investors did not want to pay software-like premium multiples before the infrastructure underpinning that growth started generating meaningful earnings.

The financial profile presented to fund managers showed that Firmus remains loss-making while it builds out its pipeline, forcing prospective shareholders to look several years ahead to an earnings base subject to severe cross-border execution risks. The key friction point is the timing of the company’s leap to the public market.

Firmus has been unusually successful at securing immense regional commitments ahead of supply, capturing a massive pipeline that exceeds 900 MW of compute commitments globally. More important, industry giants OpenAI and Meta are among the anchor customers supporting its next phase of expansion.

However, pre-selling capacity does not eliminate the physical reality of building it. Only two of Firmus’ seven planned “AI factories” are currently operational, located in Australia and Singapore. The remaining five, spread across Australia, Malaysia and Indonesia, are targeted to come online over the next two years.

For public markets, the investment case, therefore, rested not only on demand but on Firmus’ ability to convert contracts into powered, equipped and functioning capacity on time and at scale—a massive hurdle given the company’s limited operating record at such immense scale.

The IPO ran directly into a global macroeconomic headwind that is increasingly squeezing APAC infrastructure builders, especially higher global interest rates and elevated borrowing costs. Unlike a pure, asset-light software play, building grid-integrated data centres is an intensely capital-heavy endeavour. With capital no longer free, institutional due diligence has turned laser-focussed on debt loads and execution timelines.

The model’s structural weight showed up in the financial profiles circulated to prospective buyers. Firmus generated a modest $50.8 Mn in operating revenue for FY26. Targeting a $30.6 Bn valuation meant asking public markets to price the company at a staggering multiple of last year’s revenue, relying on the assumption that it would scale flawlessly to billions in earnings within a few years.

The Fundamental Metrics & Reality Gap

The transaction’s withdrawal was not triggered by a lack of fundamental interest in artificial intelligence. Instead, public equity markets drew a sharp line between future contractual promises and today’s operational earnings capacity.

Strategic Parameter	Corporate & Operational Reality
August Private Funding	Secured $2 Bn, anchoring a private valuation exceeding $10.5 Bn
October Public IPO Target	Sought to raise up to $5.5 billion targeting a $30.6 billion market cap.
Contracted Pipeline	Exceeds 900 Megawatts (MW) of global compute commitments.
Operational Infrastructure	Only 2 of 7 planned AI factories are generating active revenue (Australia and Singapore).
Under-Construction Backlog	5 regional facilities remain up to 2 years from completion across Australia, Malaysia, and Indonesia.
Short-Term Financials	Structurally loss-making start-up phase, with a projected A$77 Mn pro forma net loss in H1 FY27

Sources: Firmus Technologies corporate announcements; IPO term sheet and draft prospectus, as reported by Reuters (September–October 2026).

The valuation gap grew even more complicated due to the offer structure. Potential buyers were uneasy as existing private shareholders would have significant freedom to sell their stock shortly after the listing. This loose lock-up raised fears of a large supply overhang and selling pressure, just as new investors were being asked to accept premium pricing.

The company’s regional expansion and blue-chip anchor tenancies remain intact, though. In September, OpenAI agreed to take dedicated capacity from two planned Firmus facilities in Malaysia, while Meta committed to GPU compute capacity and long-term expansion.

The crown jewel is a planned 360 MW AI factory campus in Batam, Indonesia, being developed with Singapore-based DayOne under an eight-year Nvidia partnership. It is designed to host up to 170,000 advanced AI accelerators.

The project leverages key regional dynamics, serving as a crucial link for Singapore, which faces land and power constraints. By combining Nvidia’s complete architecture with Firmus’ unique liquid-cooled infrastructure, the Batam hub places the company at the centre of Southeast Asia’s growing digital pathways.

But the immense scale of these projects means that local partners and supply chains are deeply exposed to the venture’s valuation metrics. This financial vulnerability triggered severe stock market contagion well before Firmus ever traded. Shares in industrial conglomerate Maas Group Holdings, which owns a 3.2% equity stake in Firmus, plummeted 22.4% in a single session as IPO repricing rumours intensified, wiping out around A$517 Mn from Maas’ market value.

Maas holds an A$855 Mn electrical infrastructure order from Firmus, part of its A$1.2 Bn secured electrical work portfolio. While it quickly attributed the share-price fall to IPO speculation, the sharp equity rout showed how tightly regional industrial partners are tied to tech valuation bubbles.

After attempts to lower the offer price from A$11 to A$9 and then to A$5.50 failed to secure sufficient investor support, Firmus’ board abandoned the listing. In a statement, the board said the ASX offer terms did not reflect the business’s underlying strength and long-term growth outlook.

Firmus will now retreat to the private arena, heavily insulated by its deep-pocketed backers. Its record of successful capital raises and deep strategic backing from Nvidia gives the company credible, well-funded alternatives to an immediate public listing. What it lacks is the public valuation benchmark it sought.

However, reports suggest that the company’s public ambitions are far from dead. Firmus is considering a future NASDAQ listing in New York, though it has not formally confirmed a direct-listing structure or timetable.

Although Wall Street has historically commanded a premium multiple for high-growth AI companies, the core lesson for regional tech entrepreneurs remains unchanged.

Customer commitments and infrastructure can be signed and financed years in advance. But public markets will no longer pay the full price years in advance. When Firmus eventually returns to the public stage, investors will judge it less on promises of what it can build and more on evidence of what it has already powered up.

Also read

Firmus Tests IPO Investors With a $30.6 Bn Bet on ASEAN AI Demand