Nvidia-backed Firmus signs deal with OpenAI for Malaysia data centre capacity

Firmus signed a multi-year agreement to supply OpenAI with computing capacity from two Malaysian data centres, lifting its total contracted customer capacity above 900MW. The deal strengthens Firmus' position ahead of a rumored Australian IPO, following a fundraising valuation above $10.5 billion backed by Nvidia, Jane Street, Blackstone funds and Coatue. The agreement underscores accelerating AI infrastructure demand and intensifying competition for regional power and data-centre capacity, while raising associated electricity and water-use concerns in Malaysia.

Analysis

The relevant public-market read-through is not the customer logo but incremental evidence that scarce, powered AI capacity is becoming a strategic input rather than a commoditized hosting service. NVDA benefits if capacity commitments translate into accelerated Vera Rubin purchase orders, but the equity sensitivity is primarily to shipment timing and financing terms: a long-duration capacity contract can support a data-center developer’s funding story without creating near-term GPU revenue. The market should demand evidence of secured power, grid interconnection, and hardware delivery schedules before capitalizing the full contracted-capacity figure.

Malaysia’s bottleneck is likely to shift from land and buildings to dependable power, transmission access, water permitting, and cross-border data rules. That favors regulated grid exposure such as TENAGA.KL and well-capitalized local infrastructure owners such as YTLPOWR.KL more directly than global alternative managers; conversely, developers without contracted power may see lower-quality projects repriced despite broad AI enthusiasm. Power-policy restrictions or higher industrial tariffs would impair project returns and could force customers to seek capacity in Singapore, Australia, or the Gulf.

For BX, the implication is modest unless its funds provide additional project financing or the private valuation becomes externally marked through an IPO. A successful listing would validate AI-infrastructure asset values and potentially improve fundraising sentiment, but it also creates a public comp that can expose aggressive assumptions around utilization, depreciation, and customer concentration. GS has no investable read-through from this development absent a disclosed underwriting or advisory mandate.

Consensus appears inclined to treat contracted megawatts as equivalent to recurring, high-margin revenue. The key contrarian issue is that AI infrastructure returns can compress quickly if GPU generations shorten useful-life assumptions, customers procure directly, or utilization falls below modeled levels; the sector’s real differentiator is cost of delivered power per GPU-hour, not headline capacity.

Key Decisions for Investors

• Maintain or add NVDA on weakness rather than chase this headline; use a 3-6 month horizon tied to Rubin supply-chain disclosures and hyperscaler capex guidance. Thesis is validated by firm delivery schedules and sustained accelerator backlog commentary; reduce if major customers defer next-generation platform deployments or capex guidance rolls over.
• Establish a small 6-12 month long in TENAGA.KL or YTLPOWR.KL only after confirming tariff treatment, interconnection awards, and data-center power contracts. The asymmetric risk is regulatory intervention on electricity/water usage or tariff subsidies; avoid treating either as a pure AI proxy before those terms are visible.
• Treat any Firmus IPO as a watch item, not a pre-IPO valuation validation for BX. Require disclosure of contract tenor, take-or-pay provisions, customer concentration, committed versus available financing, power cost escalators, and depreciation assumptions; a revenue multiple based solely on contracted MW would be vulnerable to material downside.
• No incremental GS position: potential fee income from a possible transaction is immaterial to earnings and is not a tradable catalyst without a formally disclosed role.