OpenAI puts its name to frontier NVIDIA Vera Rubin compute in Malaysia through an Australian neocloud that is days from a public listing — no contract value disclosed, no megawatts attributed, no GPU count on the record.
What Happened
Confirmed first by the Firmus newsroom on September 8 and reported by The Australian, OpenAI has become an anchor customer of Firmus — the NVIDIA-, Coatue-, and Blackstone-backed Australian AI-factory operator — under a multi-year strategic partnership for dedicated compute at two Firmus AI Factory sites in Malaysia. The systems are specified as NVIDIA’s next-generation Vera Rubin NVL72 infrastructure. OpenAI’s VP of Compute Strategy, Sachin Katti, is quoted directly: the new Malaysian data centers “will help us serve growing demand for OpenAI’s products across the region and around the world.” That is the confirmed record. The first-party newsroom release outranks the paywalled headline; no second major masthead had confirmed details as of publication.
Three things to be explicit about before going further. First: no contract value has been disclosed. The Australian called it “massive” — that is the paper’s characterization, not a number anyone has put on the record, and it should be treated accordingly. Second: this is a genuinely new, separate deal. It must not be conflated with the earlier Firmus contract signed in March 2026 for a Melbourne facility involving approximately 18,400 GB300 GPUs at a reported ~A$4.5 billion — that is a different country, a different chip generation, a different (unnamed) customer, and independent reporting has pointed toward Meta, not OpenAI, as Firmus’s largest Australian anchor. Third: the “more than 900 megawatts” Firmus now cites as its total contracted capacity is its cumulative book across all customers combined — it is not the size of the OpenAI agreement, and attributing that figure to this deal would be a factual error.
Firmus is targeting an ASX listing later in 2026. Reported float valuations have ranged widely — from roughly A$5.5 billion to considerably higher — and none of those figures should be stated as settled fact. What is settled: OpenAI’s name is now attached to Malaysian frontier compute through an operator that is days from asking public markets to price it. The timing is part of the story.
The key insight: The most significant fact in this announcement is not the money — which hasn’t been disclosed — it’s the map. The same country that was reported Monday to be weighing Huawei’s Ascend silicon for its sovereign cloud now has OpenAI’s Western-stack Vera Rubin compute landing inside it through an Australian neocloud. Malaysia is not picking a side. It is becoming the place where both AI stacks physically touch down.
The Structural Read
There are two separate dynamics running through this announcement, and conflating them produces the wrong analysis.
First: Malaysia as contested compute geography. On Monday, Bloomberg reported that Malaysia is evaluating Huawei’s Ascend chips for its national sovereign AI cloud — Chinese silicon being actively courted for state-level workloads. Now, twenty-four hours later, the other side of that ledger arrives: OpenAI, the flagship of the American AI stack, is placing frontier NVIDIA Vera Rubin compute into the same jurisdiction through an Australian operator. This is not a clean bloc split. This is a hedged hub. Malaysia is not choosing between the American and Chinese AI stacks — it is taking capital and hardware from both, sometimes street-by-street in the same data-center corridors around Kuala Lumpur and Johor. That is what genuinely contested compute geography looks like in practice. Neither stack dominates; both rent space next door to each other and compete for the workloads of a government that has every incentive to stay non-aligned.
Second: OpenAI’s multi-neocloud compute sprawl. OpenAI has been systematically adding anchor-customer arrangements with a widening roster of neoclouds and AI-factory operators — a deliberate strategy to reduce single-provider dependence and extend its inference footprint into geographies where hyperscaler capacity is thin or politically constrained. Firmus is the newest name on that list, and the first to put OpenAI capacity specifically into Southeast Asia at the Vera Rubin generation. The pattern is the point: each deal is with a different operator, in a different jurisdiction, on progressively newer silicon. The compute sprawl is deliberate, not incidental.
Sachin Katti — VP Compute Strategy, OpenAI (via Firmus Newsroom, Sep 8 2026)
“[The new Malaysian data centers] will help us serve growing demand for OpenAI’s products across the region and around the world.”
The IPO-signal dimension. A named-anchor-customer announcement landing days before a public listing, with the contract value conspicuously withheld, is as much a capital-markets event as an operational one. An OpenAI logo is worth a great deal to a neocloud’s IPO book. The absence of a disclosed figure is not an oversight — it is consistent with how these pre-float anchor arrangements work: the name does the marketing, the terms stay private. Sophisticated investors will price the signal; retail investors should note what is and isn’t on the record.
Map of AI — Contested Compute Geography
The Neutral-Ground Hub Pattern
When a sovereign state refuses to align exclusively with either AI stack, it becomes a hub where both stacks physically co-locate. The winning move for that state: extract capital investment, jobs, and technology transfer from both sides simultaneously. The risk: becoming a chokepoint that either side can pressure when the geopolitical temperature rises. Malaysia, right now, is running that playbook — American inference capacity on Vera Rubin silicon next to a sovereign cloud potentially running Ascend. That is not a contradiction. It is the hedge.
Three Implications
IMPLICATION 1 — MALAYSIA BECOMES THE SWING JURISDICTION IN AI COMPUTE
With Huawei Ascend being considered for the sovereign cloud and OpenAI’s Vera Rubin compute arriving via Firmus, Malaysia is now the clearest live example of a contested compute geography — a neutral-ground hub where American and Chinese AI infrastructure physically coexist. How Malaysia regulates data residency, export-control compliance, and state-cloud procurement over the next 18 months will determine whether it becomes a stable dual-stack hub or a pressure point in a sharper US-China technology split. Neither outcome is guaranteed.
IMPLICATION 2 — OPENAI’S NEOCLOUD SPRAWL STRATEGY IS ACCELERATING AT THE VERA RUBIN GENERATION
Firmus is the first operator to put OpenAI capacity into Southeast Asia at the Vera Rubin generation — the newest chip tier, not a prior generation being recycled into a new market. That matters: it means OpenAI’s inference footprint is reaching new geographies at the frontier of compute, not a generation behind. Each new neocloud anchor deal also reduces OpenAI’s leverage risk with any single provider. The roster is widening; the silicon is current.
IMPLICATION 3 — THE ANCHOR-LOGO-AS-IPO-SIGNAL PATTERN IS NOW STRUCTURAL FOR AI INFRASTRUCTURE FLOATS
A named hyperscaler or frontier-lab anchor customer, announced days before a public listing with terms withheld, is becoming a repeatable playbook for AI-factory operators entering public markets. The OpenAI name provides credibility that no disclosed revenue figure can replicate at the pre-revenue or early-revenue stage. Investors evaluating Firmus’s float should read the anchor announcement as a demand signal and a risk-reduction narrative — and separately ask what they are not being told about term length, minimum commitments, and ramp schedules. The signal is real; the due diligence gap is equally real.
