OpenAI just anchored an Australian company’s Malaysian data centres

Australian data-centre company Firmus announced a partnership under which it will build data centres in Malaysia using U.S.-made chips and provide dedicated compute capacity to OpenAI. The deal positions Asia-Pacific as a producer rather than solely a consumer of AI computing capacity, although no contract value, chip volume, or capacity figures were disclosed.

Analysis

The investable read-through is not Firmus itself but a further broadening of AI infrastructure demand beyond US hyperscaler campuses. Incremental Asia-Pacific capacity should disproportionately benefit the equipment stack with high content per megawatt—NVDA/AMD accelerators, AVGO networking, VRT thermal management, ETN electrical distribution and GEV grid equipment—while reducing the market’s concern that AI capex is geographically concentrated or nearing a US power-permitting bottleneck. The key question is whether this is a funded, power-secured deployment rather than a capacity reservation; without disclosed MW, GPU generation, financing and utility contracts, it is not yet material enough to alter FY estimates.

Malaysia creates a potentially important second-order supply-chain node because it combines semiconductor assembly/test depth with lower power and land costs than Singapore. That can divert regional AI workloads and ancillary investment away from Singapore-listed data-center beneficiaries and toward Malaysian power, fiber and industrial infrastructure, but it also raises execution risks around grid reliability, water availability and cross-border data governance. Over 6-18 months, dedicated regional capacity could support a more resilient demand base for accelerator vendors; it is less clearly positive for US data-center REITs such as EQIX and DLR if customers can satisfy Asia demand locally rather than through Singapore or US availability zones.

Consensus may overstate the immediate benefit to NVDA: dedicated-compute announcements frequently have long construction and energization lead times, and chip orders are often conditional on financing. The nearer-term catalyst is disclosure of contracted MW, a named chip architecture and utility interconnection status; a meaningful cancellation, a delay in Malaysian power delivery, or tighter US export-control interpretation for advanced accelerators would undermine the regional-buildout thesis.

Key Decisions for Investors

• Maintain a 3-6 month overweight basket of VRT and ETN versus broader AI software exposure: each incremental AI megawatt has immediate cooling and electrical content, while regional capacity announcements reinforce backlog visibility. Use a 10-12% downside stop from entry; thesis is falsified by weaker bookings/backlog conversion or evidence that projects lack power contracts.
• Watch for disclosed Firmus committed MW and GPU purchase orders before adding NVDA. If a funded deployment specifies current-generation accelerators, buy NVDA on post-announcement weakness with a 6-12 month horizon; absent volume disclosure, treat the news as sentiment support rather than an earnings catalyst.
• Consider a 6-12 month pair trade long VRT / short EQIX only if multiple Asia-Pacific enterprise customers announce local dedicated capacity. The trade expresses infrastructure-content growth over potential regional colocation substitution; exit if EQIX reports accelerating Asia-Pacific bookings or if Firmus capacity remains pre-construction.
• Set an alert for Malaysian utility interconnection awards, renewable-power procurement and financing disclosures. These are the gating datapoints for a broader ASEAN AI-infrastructure theme; until then, avoid illiquid Malaysian infrastructure proxies where valuation and project-level economics cannot be independently verified.