Firmus turns Nvidia access into a $30 billion AI cloud bet

Firmus Technologies struck a strategic partnership with Nvidia to deploy 170,000 Nvidia GPUs in Batam, Indonesia, turning the Australian AI infrastructure builder into a reseller of Nvidia-powered cloud capacity. The deal, which includes Nvidia receiving product revenue and a share of cloud revenue, is expected to generate up to $30 billion in revenue over six years based on customer commitments. Firmus aims to level the AI compute playing field for smaller companies by providing access to Nvidia hardware that would otherwise require hyperscaler-level credit ratings.

Firmus Technologies co-CEO Tim Rosenfield has struck a strategic partnership with Nvidia https://www.nvidia.com/?ref=runtimewire that would put 170,000 Nvidia GPUs in Batam, Indonesia, and turn the Australian-founded AI infrastructure builder into a reseller of Nvidia-powered cloud capacity for AI companies trying to compete without hyperscaler balance sheets, Reuters reported https://www.reuters.com/world/asia-pacific/australias-firmus-technologies-strikes-ai-access-deal-with-nvidia-2026-06-28/?ref=runtimewire . The deal is both a supply agreement and a financing signal. Firmus will buy Nvidia infrastructure and sell Nvidia-powered cloud services to "AI Native" customers and others, with Nvidia receiving product revenue and a share of Firmus cloud revenue, according to Reuters. Firmus told Reuters the deployment is expected to arrive from the first quarter of 2027 through the start of 2028, and that Firmus expects up to $30 billion in revenue over the first six years of the arrangement, based on customer commitments. That last phrase does most of the work. Firmus has not named the customers behind those commitments, and the six-year revenue figure is a company projection, not reported revenue. But even as a target, the number shows what Rosenfield and co-CEO Oliver Curtis are selling to investors and customers: not another GPU rack operator, but a financing and distribution layer for AI compute in a market where access to Nvidia hardware has become a credit problem as much as an engineering problem. "We have worked to figure out how to close the gap between the cost benefits that the large guys have access to, which they do because they have great credit ratings, and the guys that are up and comers," Rosenfield told Reuters. "This is actually a really material way to level the playing field a little bit to give the next a chance to compete with the big guys." From factory design to regional scale Firmus is not arriving at this from a standing start. On its website, Firmus https://firmus.co/about?ref=runtimewire says it was incorporated in Australia in 2019, completed early HyperCube research and development in 2020, and had a sovereign AI services platform live in Singapore by 2023. In April, Reuters reported that Firmus had raised $1.35 billion over the previous six months at a $5.5 billion post-money valuation. The new Batam deal is separate from Firmus' Australian Project Southgate push, but it follows the same thesis: turn energy-aware infrastructure into lower-cost AI capacity. Firmus says its platform is built around high-efficiency AI factories, with cooling, power, grid orchestration and compute designed together rather than bolted on after the data center shell is built. On its website, Firmus says https://firmus.co/about?ref=runtimewire it was already designing systems for more than 1,000 GPUs in 2019 and has built an AI Factory platform spanning cooling systems and grid orchestration software. Nvidia is not just the vendor The Nvidia relationship is deeper than a purchase order. Reuters reported that Nvidia participated in prior Firmus capital raises, according to Firmus, making Nvidia an investor. Firmus also said in September 2025 that it closed a US$330 million equity placement with a cornerstone investment by Ellerston Capital and participation from Nvidia, with the proceeds aimed at accelerating Project Southgate in Tasmania. In that Firmus announcement https://firmus.co/newsroom/firmus-closes-usd330m-raise-with-nvidia-joining-as-an-investor?ref=runtimewire , Rosenfield said Firmus' mission was to create "the most efficient AI infrastructure," and Firmus said Project Southgate would include 36,000 Nvidia GPUs built over two stages. That structure gives Nvidia multiple ways to win. Nvidia sells the infrastructure, holds an investor position in Firmus, and, under the new Reuters-reported agreement, participates in cloud revenue. For Firmus, Nvidia's backing is a credibility instrument with customers and capital providers. For Nvidia, Firmus becomes another channel to move advanced AI infrastructure into markets and customers that may not have the scale, credit profile or procurement leverage of the largest model labs and hyperscalers. It also makes Firmus part of a broader shift in AI infrastructure, where Nvidia-backed or Nvidia-aligned cloud providers are trying to intermediate access to GPUs for customers that cannot wait for traditional cloud capacity or cannot justify owning the hardware outright. The competitive field includes hyperscalers, neoclouds and regional AI infrastructure operators. Firmus' differentiation, as Firmus tells the market, is not just the GPUs. Firmus is pitching a full-stack build that ties power, cooling, network, software and financing into one operating model. Batam gives Firmus a regional compute beachhead The Batam location is the most important geography in the Reuters story. Batam sits near Singapore, one of Asia's core connectivity and data center hubs, but Reuters did not report details on the Batam facility's power supply, cooling design, permitting, ownership or construction partners. Those missing facts matter because 170,000 GPUs are not a software deployment. They require power availability, grid stability, water or liquid-cooling infrastructure, networking, sovereign and cross-border data considerations, and a construction schedule that leaves little room for delay if capacity is supposed to arrive between early 2027 and early 2028. Firmus has spent the last year laying out a similar argument in Australia. In March, Firmus said it signed a long-term contract with a leading global technology company for dedicated AI infrastructure capacity at Project Southgate's first Australian deployment, covering approximately 18,400 Nvidia GB300 GPUs in Melbourne. In the same Firmus statement https://firmus.co/newsroom/firmus-signs-multi-year-agreement-with-global-hyperscale-customer-at-project-southgate?ref=runtimewire , Firmus said Project Southgate's broader buildout was backed by a US$10 billion debt facility led by Blackstone https://www.blackstone.com/?ref=runtimewire and that a Tasmania campus would host approximately 36,800 Nvidia GB300 GPUs when completed in late 2026. Firmus has also described Project Southgate as a national initiative scaling to 1.6 gigawatts by 2028, with deployments across Australia and a flagship green AI campus in Tasmania. A Firmus partnership announcement with Megaport https://firmus.co/newsroom/firmus-project-southgate-goes-on-net-with-megaport?ref=runtimewire said Southgate's first stage was a $4.5 billion investment in Australia's AI capabilities and that an initial Melbourne cluster of 18,500 Nvidia GB300 GPUs would be live in Q1 2026. Those company statements are ambitious, and they are company statements. But they explain why the Batam agreement fits Firmus' pattern. Rosenfield and Curtis are not building a single site. They are building a regional AI infrastructure finance-and-delivery machine, one that tries to match Nvidia supply, customer commitments and large-scale capital into repeatable GPU campuses. The IPO pressure is already visible Reuters reported that Firmus has appointed investment banks to work on a potential initial public offering, citing people familiar with the matter. Rosenfield declined to comment to Reuters on IPO preparations. Reuters also said Firmus had raised $1.35 billion over the previous six months at a $5.5 billion post-money valuation. That financing backdrop raises the stakes for the Nvidia partnership. A potential listing story needs visible growth, contracted demand, strategic suppliers and a credible answer to the capital intensity of AI data centers. The Reuters-reported $30 billion six-year revenue expectation gives Firmus a headline growth frame. The unverified customer commitments behind it, the delivery timing, and the economics of Nvidia's cloud revenue share will determine whether that frame survives investor diligence. For Rosenfield, the bet is straightforward and difficult: use Nvidia's supply relationship and Firmus' energy-aware infrastructure pitch to make GPU access cheaper for customers that are not OpenAI-scale, Google-scale or Microsoft-scale. If Firmus can deliver Batam on schedule, the deal gives emerging AI companies another path to compute. If Firmus cannot, the partnership becomes another reminder that in AI infrastructure, access is only the first bottleneck. Execution is the harder one.