AI Bubble Fears Mount as DayOne Weighs IPO Cut, Firmus Turns to Nvidia
Singapore’s DayOne is weighing a smaller or deferred U.S. listing, and Australia’s Firmus is in talks with Nvidia for fresh funding after pulling its IPO. A rout in data-centre stocks is testing how far investors will finance AI’s infrastructure ambitions.
Opinions expressed by Entrepreneur contributors are their own.
You're reading Entrepreneur Asia Pacific, an international franchise of Entrepreneur Media.
“There’s almost certainly bubbles within this moment in time. And any radical transformation of our economy is generally intertwined with a bubble,” Ken Griffin, the founder of the U.S. hedge fund Citadel, recently told a conference in New York.
At the Future of Global Markets conference, hosted by Citadel Securities, his market-making firm, Griffin compared the current AI boom to the railway boom and the dot-com era. Both brought significant economic changes but also fuelled speculative excess. Griffin cautioned that while AI is unlikely to fail, excessive enthusiasm could lead investors to support inflated valuations and uncertain returns.
In Asia-Pacific, companies building AI’s physical infrastructure are already seeing what that means for their funding.
Singapore’s DayOne Data Centers, which has 962 MW of data centre capacity in service and sites across 10 markets in Asia-Pacific and Europe, is reportedly considering halving its planned U.S. IPO or shelving it for now. Nvidia-backed Firmus of Australia, an AI computing provider, has already pulled a listing that would have raised nearly $5 Bn and is reportedly in talks with the chipmaker about fresh funding.
Meanwhile, Australian data centre operator NextDC lost 41.7% of its market value in a single session after a delay to grid upgrades at one of its Melbourne sites forced a sharp cut to its earnings forecasts.
Each setback is structurally different, but investors’ message remains loud and clear. They are no longer willing to pay up front for AI capacity that has yet to produce earnings, whether the hurdle is valuation, financing or delivery.
DayOne’s $5 Bn IPO Dilemma
DayOne, which has built most of its data centre business in Johor, Malaysia, may cut its U.S. IPO from as much as $5 Bn to $2.5–$3 Bn, Bloomberg reported on October 21, citing people familiar with the matter. It is also weighing a postponement until markets settle, although it remains in discussions with prospective investors and has not made a final decision.
The rethink comes soon after DayOne filed for a NASDAQ listing under the ticker DODC on October 5. Its prospectus outlines a rapidly growing business, but its development programme will require billions more.
Its revenue rose 238% YoY to $512 Mn in the first half of 2026, up from $151.5 Mn, driven by new operational capacity in Malaysia, Indonesia and Thailand. DayOne capacity billed to customers climbed from 213 MW in June 2025 to 666 MW a year later, showing the expansion is already generating revenue. That figure passed 1 GW in mid-October.
Losses, however, grew faster. The net loss widened to $77.2 Mn from $12.6 Mn, driven by a one-time charge to terminate a customer support arrangement, share-based payments and currency headwinds.
The more pressing issue is funding. DayOne carried $4.9 Bn in borrowings and finance leases as of June. It holds about 2.3 GW of contracted bookings, nearly all due for delivery by the end of 2028, and estimates it will need another $11.4 Bn to complete them.
The company says its available capital resources should be adequate to fund those bookings. It raised about $3.2 Bn in Series C in 2026 and recently upsized its Malaysia bank financing to about $7.2 Bn. Hence, a smaller or delayed IPO would only reflect what public investors are willing to pay, not DayOne’s capacity to build.
Customer concentration further complicates that valuation. DayOne’s largest customer, described in the prospectus as a global technology company with a leading short-form video platform, generated 69.2% of first-half revenue, and the second-largest accounted for another 15.1%. Contracts of that scale offer revenue visibility and ease project financing, but they tie DayOne’s fortunes directly to the capital expenditure plans of just two companies.
DayOne is not lagging in commercial momentum. Its immediate challenge is convincing the market that its growth rate justifies the near-$20 Bn valuation it has reportedly sought, despite its funding needs and customer exposure.
Bloomberg Intelligence analyst Stephen Yin argues that a smaller raise at a lower valuation would keep the order book stable while funding DayOne’s construction pipeline.
Firmus: When Public Investors Say No
Firmus faced a sharper rejection recently. The Australian AI computing provider pulled its listing on October 9 after failing to secure sufficient institutional support for an offering that would have raised about $5 Bn at a $30.6 Bn valuation.
That target was nearly three times the $10.5 Bn-plus valuation set just two months earlier, when Firmus announced a fully subscribed $2 Bn equity round backed by Nvidia, Coatue, Blackstone-managed funds and Jane Street.
Investors questioned whether the pricing adequately reflected the structural risks of building physical facilities, financing expensive silicon and converting customer commitments into cash earnings. The offering’s structure, which would have permitted existing shareholders to exit, also drew scrutiny.
Firmus had two AI factories operating in Australia and Singapore, and five under development in Australia, Indonesia and Malaysia, according to its September disclosures. In withdrawing its IPO, it stated market volatility meant the offering could not proceed on terms that reflected its long-term prospects. Bloomberg reported the same day that the company was exploring a private capital raise of $2–3 Bn.
Nvidia may now step in to supply some of the capital public investors withheld. The Information reported on October 20 that the chipmaker was in talks to invest hundreds of millions of dollars in Firmus, although no formal terms or valuations have been disclosed.
An extra shot of Nvidia funding may allow Firmus to keep building without returning to public equity markets, buying it time to prove it can turn contracts into revenue. But it won’t fix the valuation discrepancy that sank the IPO.
Private capital comes with its own pricing terms, investor protections and dilution risks, and there is no basis yet to assume Nvidia will endorse the valuation public markets rejected.
Demand itself is not the bottleneck. Firmus stated in September that its contracted capacity had passed 900 MW, including agreements with OpenAI for two AI factories in Malaysia, while Meta also holds infrastructure arrangements with the business. This week, Firmus named Thinking Machines Lab, the AI start-up founded by former OpenAI chief technology officer Mira Murati, as an anchor customer for its Asian AI cloud region.
Contracts, however, are not earnings. Fund managers pointed out that Firmus had only about 46 MW operating against more than 900 MW under contract. Wells Fargo estimates that building 900 MW of compute would take about $41 Bn of capital spending.
Neoclouds, which build and run large clusters of AI processors, must spend heavily on equipment and infrastructure long before the bulk of their expected revenue arrives. Private funding may give Firmus operational breathing room, but it does not diminish the sheer cost of building out its contracted capacity.
AI Infrastructure IPOs under Pressure
| Metric | DayOne Data Centers (Singapore) | Firmus (Australia) |
| Proposed IPO Size | Up to $5 Bn | Approximately $5 Bn |
| Revised IPO Size | Considering $2.5–3 Bn | IPO withdrawn on October 9, 2026 |
| Reported IPO Valuation | Approximately $20 Bn | $30.6 Bn |
| Earlier Private Valuation | Not disclosed on a comparable basis | More than $10.5 Bn (August 2026) |
| Latest Financing Development | Considering a smaller or delayed NASDAQ listing; upsized Malaysia bank financing to about $7.2 Bn | Exploring private financing; Nvidia reportedly in talks to invest hundreds of millions of dollars |
Sources: Bloomberg, The Information, Firmus company announcements (August 7 and October 9, 2026), DayOne announcement (October 14, 2026)
Note: All figures are in US dollars. IPO size represents proposed fundraising proceeds, not company valuation. Valuations are reported or proposed figures, not final public-market valuations. Neither IPO has been completed.
Nvidia’s Role Goes Beyond Funding
Nvidia is already a strategic investor in Firmus and the primary supplier of its hardware ecosystem. In June, the two announced a computing partnership running through to 2034, centred on a planned 360 MW AI factory campus on the Indonesian island of Batam, just off Singapore.
The agreement covers up to 170,000 Nvidia accelerators to be deployed through 2027 and 2028.
DayOne is developing that exact campus, linking the two companies, both of which currently face distinct forms of public-market resistance. DayOne builds and operates the underlying data centres, while Firmus sells the raw computing power running inside them.
Under the partnership, Nvidia provides credit support for equipment purchases and earns both hardware revenue and a share of the cloud revenue from the supported capacity. Based on customer commitments, Firmus expects to receive $25–30 Bn from offtake agreements over the first six years. That is money customers have agreed to pay as the Batam capacity comes online from early 2027, not revenue Firmus has booked or cash it has collected. For scale, Firmus reported just $51 Mn in revenue in FY26.
It shows how intertwined AI financing has become. Nvidia sells more silicon when its customers can afford to build, so financing those customers directly supports its demand pipeline. This deepens the structural co-dependence between supplier and buyer, a dynamic that a fresh equity investment in Firmus would only intensify, possibly on terms different from its August round.
Neither DayOne nor Firmus has said that its IPO difficulties have affected the Batam campus. The broader structural question is whether primary suppliers such as Nvidia will become a more significant source of capital for infrastructure developers as public equity markets grow less accommodating.
NextDC’s 41.7% Collapse Exposes Grid Risks
NextDC’s market rout had nothing to do with a listing debut. The established Australian operator shed 41.7% of its equity value on October 21, erasing roughly A$3 Bn in market capitalisation, after slashing its guidance for FY27. The stock rebounded 7.5% the next day but slipped again on October 23. Morgan Stanley, which downgraded NextDC to equal-weight after the guidance cut, slashed its price target to A$6.70 from A$19.
The operational bottleneck was grid power. Electricity distributor Jemena suspended parts of the critical grid infrastructure upgrades needed to increase power supply to NextDC’s M3 Melbourne facility, citing local community and stakeholder opposition and a lack of local council consent. Until that work is finished, NextDC cannot deliver capacity customers have already signed up for, or start billing for it. Jemena does not expect to resume the work before February 2027.
Consequently, NextDC cut its 2027 net revenue forecast down to A$475–A$500 Mn from an initial estimate of A$615–A$640 Mn. It also lowered its underlying EBITDA guidance to A$245–A$265 Mn, down from A$385–A$410 Mn.
The revised forecasts exclude about A$140 Mn each of net revenue and underlying EBITDA from the affected M3 contracts, including capacity already being billed, because NextDC said it could no longer reliably estimate their contribution.
In response, management lowered its capital expenditure guidance to A$4.5–5 Bn from A$5.25–5.75 Bn. Because the new range includes about A$230 Mn of capitalised interest that the earlier one left out, the like-for-like cut is about A$980 Mn, or 17–19%. NextDC also paused all uncontracted development and land acquisitions pending a comprehensive review of its capital programme.
The review comes five months after NextDC opened KL1 in Kuala Lumpur, its first data centre outside Australia, with 65 MW of planned capacity.
The affected capacity remains under contract, and construction of the data halls continues, but the delay is more than a matter of timing. The contracts provide for service-level credits when capacity is late or not fully powered and, in some circumstances, give the customer the right to terminate. NextDC warned that such credits could push underlying EBITDA below its revised range.
Long-term offtake agreements won’t matter until baseline power, municipal permits and structural construction work together to enable capacity building. NextDC’s sudden collapse underscores how rapidly the mismatch between contracted bookings and operational execution can impair public earnings.
Capital Flows Continue, but Terms are Shifting
None of these friction points indicates that institutional capital has dried up. Mergers and acquisitions targeting Asian data centres surged to $21.3 Bn so far in 2026, more than double the $10 Bn in deal value recorded in 2025, according to Dealogic data Reuters cited on October 15. Furthermore, regional syndicated loans and corporate bonds raised $22.3 Bn this year, while public equity offerings drew $5.4 Bn, down 4.6% from the same period in 2025.
Capital Flows into Asian Data Centres
| Financing Category | 2026 Value ($ Bn) | Growth/Slump |
| M&A | 21.3 | More than twice the $10 Bn recorded in 2025 |
| Syndicated Loans & Corporate Bonds | 22.3 | N/A |
| Equity Capital | 5.4 | Down 4.6% YoY |
Source: Dealogic data, as reported by Reuters (October 2026).
Note: Figures cover January 1–October 13, 2026. The M&A comparison is against full-year 2025, while the equity fundraising comparison is against the corresponding period of 2025. Categories represent different types of capital activity and should not be aggregated as total new investment.
The divergence shows different capital tiers are pricing structural risk more closely. Infrastructure lenders can comfortably rely on visible contracted income, investment-grade credit profiles and physical, completed assets. Conversely, public equity investors must weigh the compounding threats of share dilution, escalating borrowing costs and long backlogs of unbuilt projects, risks that intensify when developers seek billions in upfront funding before their capacity yields cash.
These corporate setbacks do not signal an end to the AI infrastructure super-cycle. Instead, they show the market is no longer willing to treat headline capacity announcements, customer commitments or long-term projected revenue as a substitute for near-term operating performance.
For ASEAN, the economic stakes are incredibly high. Singapore has firmly established itself as a leading financing, legal and corporate base for regional digital infrastructure, while neighbouring Malaysia and Indonesia continue to attract larger projects such as DayOne’s 1.4 GW of bookings in Johor and the 360 MW Batam campus it is developing for Firmus.
These major capital allocations inject vital construction, energy-grid and technology-sector employment into local economies. However, they yield viable financial returns only if facilities are completed on schedule, financed sustainably and operated at costs that deliver acceptable shareholder margins.
If public equity valuations remain depressed, developers may pivot to private equity, strategic joint ventures or downsized listings. Or they may freeze development until they can show substantial operational capacity and cash earnings.
Nvidia’s ongoing discussions with Firmus suggest that hardware suppliers may step in where public markets will not pay the price existing shareholders want. Yet strategic corporate money does not eliminate the risks of construction delays, rising leverage costs or compressed margins, nor does it inherently validate prior speculative valuations.
Ken Griffin’s warning frames a much narrower problem than whether AI will eventually transform the broader macroeconomy. The underlying technology may continue to expand exponentially for years. What is being stress-tested now is the financial architecture supporting it.
For DayOne, Firmus and their immediate peers, future institutional viability depends less on the total gigawatts they announce and more on their practical ability to deliver infrastructure, generate free cash flow and justify the high entry price investors are being asked to pay.
The artificial intelligence revolution may be in its infancy, but the market’s willingness to fund every ambitious blueprint at a speculative valuation is wearing thin.
“There’s almost certainly bubbles within this moment in time. And any radical transformation of our economy is generally intertwined with a bubble,” Ken Griffin, the founder of the U.S. hedge fund Citadel, recently told a conference in New York.
At the Future of Global Markets conference, hosted by Citadel Securities, his market-making firm, Griffin compared the current AI boom to the railway boom and the dot-com era. Both brought significant economic changes but also fuelled speculative excess. Griffin cautioned that while AI is unlikely to fail, excessive enthusiasm could lead investors to support inflated valuations and uncertain returns.
In Asia-Pacific, companies building AI’s physical infrastructure are already seeing what that means for their funding.