Firmus secures OpenAI Malaysia compute deal to anchor 900MW expansion

The agreement secures a high-profile anchor tenant for Firmus as it scales infrastructure across Southeast Asia, providing the revenue visibility needed for a planned 2026 IPO. By deploying Nvidia Vera Rubin systems with specialized liquid cooling, Firmus is positioning Malaysia as a critical hub for the heavy compute required by generative AI models. This shift signals that the physical location of AI processing is moving toward regions with lower power costs and high connectivity, potentially lowering the long-term operational overhead for AI-integrated streaming services. Watch for Firmus to finalize its 360 MW Indonesian campus as a benchmark for regional compute density.

Firmus Technologies is part of a broader wave of AI infrastructure providers racing to secure hyperscale compute capacity in Southeast Asia, where lower power costs and government incentives are attracting major deployments. Nvidia's Vera Rubin NVL72 platform, which Firmus will deploy in its Malaysian facilities, represents the chipmaker's next-generation rack-scale architecture designed for training and inference of frontier models. Nokia and Nvidia have deepened their AI-RAN partnership with a $1 billion investment from the chipmaker, illustrating how Nvidia is simultaneously expanding its footprint across both AI compute and telecom network infrastructure in the region. The Malaysian government has actively courted data center investment through its National Energy Transition Roadmap, offering green energy guarantees and streamlined permitting that have drawn commitments from Microsoft, Google, and Amazon Web Services alongside newer entrants like Firmus.

The business model Firmus is pursuing mirrors a pattern across the AI infrastructure market where providers lock in multi-year anchor tenants to underwrite capital-intensive buildouts. Nokia combined with AWS and Databricks to build a telco AI control layer announced at DTW in June 2026, demonstrating how infrastructure vendors are stacking partnerships to create integrated platforms that justify long-term contracts. Firmus's DSX platform and HyperCube liquid-cooling technology differentiate it from commodity colocation providers by targeting the extreme power densities required by Nvidia's latest GPU racks, which can exceed 120 kW per cabinet. The company's planned IPO in 2026 will test whether public markets value AI-specific infrastructure at a premium over traditional data center operators, a question that gains urgency as OpenAI's compute demands continue to outpace supply.

Technical performance benchmarks for liquid-cooled GPU deployments remain a key differentiator as operators compete for AI workloads. Ericsson launched its AI in RAN commercial software subscription in June 2026, claiming up to 20% higher downlink throughput across more than 15 live deployments, showing how AI-driven optimization is being commercialized across network infrastructure with measurable efficiency gains. For Firmus, the relevant benchmark is power usage effectiveness at scale: its proprietary cooling system targets PUE below 1.15 in tropical climates, a figure that would place its Malaysian facilities among the most efficient in the region. , providing a parallel example of how Nvidia-backed platforms are delivering quantified performance gains across infrastructure categories. The streaming industry should watch these developments because the same GPU architectures powering AI model training increasingly serve video encoding, content recommendation, and real-time personalization workloads that underpin modern streaming platforms.