OpenAI Signs Multi-Year Malaysia Data Center Deal with Firmus

Australian AI infrastructure firm Firmus announced on September 8, 2026 that it has signed a multi-year deal with OpenAI to supply computing capacity from two data centers in Malaysia. The agreement makes OpenAI an anchor customer and pushes Firmus's total contracted capacity past 900 megawatts across its global network. The deal also sets the stage for a rumored Firmus IPO, which analysts expect could rank among Australia's largest public listings in recent years.

What happened

Detail	Fact
Announcement date	September 8, 2026
Deal type	Multi-year computing capacity agreement
Data center locations	Two sites in Malaysia
Firmus total contracted capacity	More than 900 MW (across all customers)
Firmus last valuation	Above $10.5 billion
Key Firmus backers	Nvidia, Jane Street, Blackstone funds, Coatue Management
Processors being deployed	Nvidia Vera Rubin (next-generation)
Contract value disclosed	No (Firmus declined to comment)

Firmus currently operates two AI data centers in Australia and Singapore, with five more under development across Asia-Pacific. Adding Malaysia extends that footprint into what the source describes as Southeast Asia’s fastest-growing data center market. Firmus will deploy Nvidia’s Vera Rubin processors, the successor to the current Hopper line, at scale across the region.

The deal came just days after OpenAI unveiled Astra, which the company called its most capable model to date. Larger and more capable models require substantially more compute during both training and inference, which explains why OpenAI is locking in capacity well ahead of need.

Why it matters

This deal is one more data point in a pattern worth watching: U.S. and Chinese AI groups are racing to secure power and physical data center space before demand outpaces supply. Malaysia has attracted heavy investment precisely because of available land, favorable regulation, and proximity to undersea cable routes. That growth has also drawn scrutiny over electricity and water consumption, a risk factor for anyone planning infrastructure there long-term.

For Firmus, landing OpenAI as an anchor customer does two things. It de-risks the business ahead of a potential IPO, and it signals to other hyperscalers that the company can operate at the scale these customers require. A valuation above $10.5 billion, backed by Nvidia directly, suggests the market already prices in significant future contracts.

For businesses and developers who rely on OpenAI’s API, more committed capacity in Asia-Pacific could mean lower latency for users in that region and more headroom for OpenAI to serve demand as its newer models roll out. These deals are a leading indicator of where AI product availability will improve over the next 12 to 24 months.

If you are thinking about AI integration for your business, the compute race playing out at this scale is a reminder that the underlying infrastructure shapes what products actually become available, and at what speed.

Our take

The number that stands out is 900 MW. For context, that is roughly the output of a mid-sized power station, committed entirely to running AI workloads. Firmus going from a regional player to an anchor-customer relationship with the world’s most prominent AI lab is a significant commercial milestone, not just a press release.

What we would watch: the rumored IPO. If Firmus lists publicly this year, the prospectus will contain detail on contract terms, revenue concentration, and margins that the current announcement deliberately hides. A company with a single dominant customer at 900 MW is a very different risk profile than one with that capacity spread across dozens of clients. The IPO documents will clarify which one Firmus actually is.

The Nvidia Vera Rubin chip deployment is also worth noting. Vera Rubin is not yet widely deployed commercially. Getting confirmed deployments at scale in Asia-Pacific, through a named partner, is a meaningful signal about where Nvidia’s next hardware cycle lands first. We covered related moves in Nvidia’s broader investment strategy in our look at Nvidia’s $99B investment portfolio, and the pattern is consistent: Nvidia is placing chips and capital simultaneously.

One honest caveat: neither the contract value nor the specific capacity allocation to OpenAI has been disclosed. The 900 MW figure is Firmus’s total across all customers. Do not read it as OpenAI’s committed share.

What to do about it

1. Note Malaysia as an emerging AI infrastructure hub if you are evaluating cloud or colocation providers in Asia-Pacific.
2. Watch for the Firmus IPO prospectus, it will contain the most detailed public look at AI data center economics to date.
3. If you are building products on OpenAI’s API, track Asia-Pacific region availability in OpenAI’s platform docs as new capacity comes online.
4. Factor infrastructure lead times into any AI product roadmap: deals being signed today translate to capacity available 12 to 24 months from now.

The short practical takeaway: compute availability is the real constraint on AI product timelines, and deals like this one are how the industry tries to get ahead of it.

Source: Bing News · OpenAI