Two documents were read for this piece: the Firmus company release and Capital Brief. The Australian Financial Review, which is reported to have written that the institutional bookbuild was pushed back by two days, is paywalled and was not read here; Reuters was not read either. Nothing below draws on either outlet, and no schedule is asserted as final.
Firmus and Meta have announced what the release describes as “strategic AI infrastructure agreements for GPU compute capacity at Firmus AI factories in Southeast Asia built on NVIDIA’s full-stack DSX platform,” with the agreements said to “include contracted capacity and additional capacity expansion options.” The release names the counterparty (Meta), the platform (NVIDIA DSX), the region (Southeast Asia), the structure (contracted plus expansion options), and two executives: Gaya Nagarajan, Vice President of Engineering and Infrastructure at Meta, who called Firmus “a long-term strategic infrastructure partner,” and Tim Rosenfield, Co-Founder and Co-CEO of Firmus. The release states no dollar amount, no megawatt or gigawatt capacity figure, and no contract length anywhere on the page. As Capital Brief puts it directly: “Neither company disclosed the financial terms of the agreement.”
What the release does quantify is the hardware. It names NVIDIA GB300 NVL72 systems at Firmus’s Melbourne AI factory — which Firmus describes, in its own words, as “the largest deployment of NVIDIA Blackwell Ultra systems in the Southern Hemisphere,” a company claim attributed rather than independently verified here. A HyperCube Manufacturing Facility in regional New South Wales is also named. The Southeast Asian sites the agreements actually cover are not named in the release.
The key insight: Specificity has been spent on the silicon and withheld from the economics. The release identifies the chip generation with precision while leaving every financial dimension of the underlying agreement undisclosed. That is an observation about the document — pre-listing disclosure is governed by rules this publication has not examined, and no impropriety is suggested.
Capital Brief describes FY28 as the year Firmus expects all of its planned AI factory sites to be operational. The uplift inherits every assumption in the forecast it is a percentage of.
The Structural Read
The deal is fully described and entirely unquantified — and into that gap, a number has arrived from outside the company. Two people with knowledge of the matter told Capital Brief that the agreement is expected to lift Firmus’s group EBIT by around 15%. The crucial detail is what that percentage is measured against: it is 15% of the USD 5 billion figure the company presented to prospective investors, producing an implied uplift of roughly USD 750 million and taking Firmus’s FY28 EBIT to around USD 5.75 billion. Capital Brief describes FY28 as the year “when Firmus expects all of its planned AI factory sites to be operational.” Neither company has confirmed any of those figures.
The structural point is worth holding carefully. The USD 750 million figure is not a contract value. It is a share of a projection for a fiscal year two years away, conditioned on a build-out that is not finished. That means the uplift inherits every assumption embedded in the forecast it is a percentage of — site timelines, utilization rates, pricing, and the operating leverage of AI factories that do not yet all exist. A deal of undisclosed size is being priced by its effect on a forward model, and the forward model carries its own uncertainty.
Map of AI — Infrastructure Layer
“A company at the infrastructure layer of the AI stack — building and operating physical compute capacity — monetizes through contracted utilization. The economics are real but the disclosure arithmetic here runs: undisclosed contract terms, times a percentage uplift, times a two-year-out forecast, equals a number that inherits three compounding layers of assumption.”
On this publication’s arithmetic, around USD 50 billion against around USD 5.75 billion of FY28 EBIT is roughly 8.7 times — a figure that divides a pitched valuation by a projected earnings number and is therefore not a market multiple. It is a ratio of two disclosed inputs, one of which is a company investor pitch and the other a forecast from unnamed sources. Capital Brief, as read, carries a bookbuild on 6 October and an ASX listing on 22 October; a later bookbuild date is reported by the AFR, which was not read here, so neither schedule is asserted as final.
Three Implications
THE HARDWARE ANCHOR
Naming NVIDIA’s GB300 NVL72 — the specific chip generation — before naming any financial term is a deliberate ordering. In infrastructure deals ahead of a public listing, the silicon spec functions as a credibility signal when commercial terms are not disclosable or not yet disclosed. The hardware is the quotable fact; the economics travel separately.
FORECAST AS PRICING SURFACE
When deal terms are undisclosed, the market reaches for whatever number is available. Here that number is a FY28 EBIT projection — conditioned on all planned AI factory sites being operational by then — that two unnamed sources have converted into an implied uplift. The risk is that the projection, not the contract, becomes the reference point investors anchor to.
SOUTHEAST ASIA AS UNNAMED CONTEXT
The agreements cover Southeast Asian AI factory sites that the release does not name. The named sites — Melbourne and regional New South Wales — are in Australia. The geography of the actual contracted capacity therefore sits outside the disclosure, which makes the Southeast Asia framing structural context rather than locatable commitment at this stage of public information.
Business Engineer Framework
The Map of AI — Infrastructure Layer
Firmus sits squarely in the physical infrastructure layer of the AI stack — the layer that owns the land, the power, the cooling, and increasingly the contracted chip capacity that every model above it depends on. The Map of AI traces how value accrues differently at each of the nine layers, and why infrastructure players monetize through utilization contracts rather than software margins. Understanding where Firmus sits in that stack is the right starting point for reading any deal announcement that names the chip and withholds the price.
The Firmus–Meta announcement is structurally interesting precisely because of what it withholds: a named hyperscaler, a named chip generation, a named region, and a named structure sit alongside zero disclosed financial terms — and the only number circulating publicly is a projection for a fiscal year two years away, sourced from two unnamed people, conditioned on a build-out that is not yet complete, and not confirmed by either company. That is the document as it exists; what the underlying contract contains is a separate question the release does not answer.
Sources: Firmus company release; Capital Brief. The Australian Financial Review is paywalled and was not read. Reuters was not read. Nothing in this article is investment advice.
Two documents were read for this piece: the Firmus company release of 29 September 2026 and Capital Brief of the same date. The Australian Financial Review, which is reported to have written that the institutional bookbuild was pushed back by two days, is paywalled and was not read here; Reuters was not read either. This piece therefore reports the 6 October bookbuild and 22 October listing dates that appear in what it read, notes that a later bookbuild date is reported elsewhere, and asserts neither schedule as final.Neither Firmus nor Meta disclosed the financial terms of the agreements. The figure of roughly USD 750 million comes from two unnamed people via Capital Brief and is not a contract value: it is 15 per cent of a USD 5 billion FY28 group EBIT figure the company presented to prospective investors, for a fiscal year Capital Brief describes as the one in which Firmus expects all of its planned AI factory sites to be operational. The uplift therefore inherits every assumption inside that forecast, and neither company has confirmed it. Where this piece observes that the release states no dollar amount, no capacity figure and no contract length, that describes the contents of the document rather than proving that no such terms exist. The description of the Melbourne installation as the largest deployment of NVIDIA Blackwell Ultra systems in the Southern Hemisphere is Firmus’ own claim and has not been independently verified here. The figure of roughly 8.7 times is this publication’s arithmetic, dividing a pitched valuation of around USD 50 billion by a projected FY28 EBIT of around USD 5.75 billion. It is not a market multiple. No raise size, revenue figure, current EBIT, megawatt or gigawatt figure, contract length or named Southeast Asian site was read, and none appears above. Nothing above predicts anything about the listing, the share price, Firmus, Meta or NVIDIA; nothing above characterises the disclosure as inadequate or suggests any impropriety; and nothing here is investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
