Firmus and Nvidia Strike a $30 Billion AI Compute Deal

Big AI labs get cheap compute because they have great credit. An Australian startup just signed a deal with Nvidia to hand that same edge to everyone else.
Key Takeaways
- 1Firmus Technologies signed a strategic partnership with Nvidia to resell Nvidia-powered cloud to emerging *AI Native* customers at lower cost.
- 2The deal deploys 170,000 GPUs in Batam, Indonesia between early 2027 and early 2028 and targets up to 30 billion dollars in revenue over six years.
- 3Nvidia is already a Firmus investor, and Firmus, valued at 5.5 billion dollars, has appointed banks to work on a potential IPO.
Australian AI infrastructure company Firmus Technologies signed a strategic partnership with Nvidia on June 28, 2026 to give emerging AI firms cheaper access to computing power.
According to Reuters, the deal lets Firmus buy Nvidia infrastructure and sell Nvidia-powered cloud services to AI Native customers, earning the chip giant product revenue plus a share of cloud revenue.
The Deal Targets the Companies the Big Labs Outspend
The premise is a cost gap. Reuters quoted Firmus co-chief executive Tim Rosenfield saying the company worked to close the gap between the cost benefits large players get from strong credit ratings and the terms available to up-and-comers.
Rosenfield called it a material way to level the playing field so the next wave of companies can compete with the incumbents. The framing positions Firmus as an access provider, not just another cloud reseller.
That gap is real. Compute can consume 40 to 60 percent of an AI startup's technical budget in its early years, a pressure the wider market has watched reshape AI compute economics.
The Numbers Behind the Agreement
The scale is substantial. Business Recorder reported that Firmus expects to earn up to 30 billion dollars in revenue during the first six years of the deal, based on customer commitments.
The hardware follows a clear timeline. Reuters reported the deal will deliver 170,000 GPUs from the first quarter of 2027 to the start of 2028, located in Batam, Indonesia.
The chips themselves are Nvidia's newest. Crypto Briefing reported that the flagship site under the broader Project Southgate effort is set to house roughly 18,400 to 18,500 Nvidia GB300 GPUs, with rollouts beginning in 2026.
Nvidia Is Both Supplier and Investor
The relationship runs deeper than a single contract. Reuters reported that Nvidia has participated in Firmus' previous capital raisings, making it an investor in the Australian firm.
Firmus has raised heavily to fund the buildout. Crypto Briefing reported that the company closed a 505 million dollar equity round at a 5.5 billion dollar post-money valuation in April 2026, led by Coatue Management with Nvidia participating, bringing total equity raised to 1.35 billion dollars.
On top of that, Firmus secured a 10 billion dollar debt facility earlier in 2026, taking combined financing past 11 billion dollars. The arrangement extends Nvidia's reach into the cloud layer, the kind of integration that runs alongside its work on tools like NVIDIA NeMo Guardrails.
An IPO May Be Next
The deal lands as Firmus weighs going public. Reuters reported that the company has appointed investment banks to work on a potential initial public offering, citing people familiar with the matter, though Rosenfield declined to comment on those preparations.
A confirmed analyst snapshot underlined the market backdrop. GuruFocus noted Nvidia's price-to-earnings ratio sat at 29.48 at the time, with significant insider selling over the prior three months.
Why It Matters for Builders
The deal is one more sign that the scramble for compute capacity now drives the AI business as much as model quality does. The same demand pressure has pushed hyperscalers to chase ever-larger buildouts, a race documented in coverage of AI infrastructure capacity.
For smaller AI companies, a GPU-rich challenger pricing against the hyperscalers is worth watching. If Firmus delivers on its access pitch, it could give early-stage teams a path to the compute they need without the credit terms only giants enjoy.
What Changed
A second-tier cloud provider locked in a supply-and-resale agreement with Nvidia aimed squarely at smaller AI firms. Firmus buys Nvidia infrastructure and sells Nvidia-powered cloud, giving Nvidia product revenue plus a share of cloud revenue.
The pitch is access. Firmus says the structure lets up-and-coming AI companies reach infrastructure on terms closer to what the largest players already enjoy.
Why It Matters
Compute cost is the single biggest constraint on early-stage AI companies, often eating 40 to 60 percent of technical budgets. A cheaper, GPU-rich cloud aimed at that tier could shift where smaller AI products get built.
It also deepens Nvidia's reach into the cloud layer through a partner it has funded, extending its grip on the AI supply chain beyond chip sales.
Suggested Actions
If you run an AI product, model compute as a first-class cost line and compare neocloud providers against the hyperscalers on price and availability. Watch GPU-rich challengers like Firmus as alternatives that can ease the margin squeeze on inference.
Related Tags
Related News
GitHub Copilot Billing Shock Hits as First Metered Cycle Closes
By Faisal Irfan
June 30 marks the close of GitHub Copilot's first full metered billing cycle, and developers running agentic workflows report costs 10x to 50x higher than their old flat-rate plans.
Microsoft Ends Claude Code Access for Thousands of Engineers on June 30 Deadline
By Muhammad Musa
Microsoft's June 30 deadline to end Claude Code access across its Experiences and Devices division has arrived, forcing thousands of engineers onto GitHub Copilot CLI.
Gartner Predicts AI Coding Costs Will Exceed Developer Salaries by 2028
By Waqas Arshad
Gartner forecasts that AI coding token costs will overtake the average developer's salary by 2028, and the evidence is already arriving from Slash, Uber, and Microsoft.






