AI Power Shift: OpenAI Secures Malaysian Compute While Anthropic Walks Away From $6B Deal

OpenAI has just cemented a multi‑year agreement with Malaysia’s Firmus to lock down hundreds of megawatts of AI‑grade compute.

At the same time, Anthropic unexpectedly walked away from a $6 billion takeover of Decart AI, signaling a shift in how start‑ups approach big‑ticket acquisitions.

The two stories, reported by Moneycontrol and CNBC TV18, illustrate the divergent strategies AI developers are adopting as the compute arms race intensifies. OpenAI’s deal with the Nvidia‑backed Firmus marks a decisive move into Southeast Asia, where data‑center capacity is expanding faster than in many Western markets. Firmus will provide OpenAI with dedicated power exceeding 900 MW across two Malaysian sites, turning the country into a new hub for the training of large language models. This partnership not only guarantees the compute horsepower required for next‑generation models but also showcases OpenAI’s willingness to lock in long‑term infrastructure contracts to hedge against volatile GPU pricing and supply constraints.

According to Moneycontrol, the agreement makes OpenAI an “anchor customer,” a status that typically grants favorable pricing, priority access to hardware, and the ability to shape data‑center specifications. The firm’s commitment also benefits Firmus, which gains a stable revenue stream and credibility that can attract other AI players looking for reliable capacity in the region. The deal underscores a broader trend: leading AI labs are increasingly looking beyond traditional cloud providers, preferring tailored, high‑density facilities that can deliver the low‑latency, high‑throughput environment necessary for training models that now exceed hundreds of billions of parameters.

Meanwhile, Anthropic’s decision to walk away from the Decart acquisition, as noted by CNBC TV18, sends a clear message about the limits of aggressive M&A in the AI sector. The $6 billion price tag would have been one of the largest AI‑focused deals to date, and Decart’s technology promised to accelerate Anthropic’s roadmap for the Claude series. However, sources suggest that the deal stalled over valuation disagreements and concerns about integrating Decart’s hardware‑centric assets into Anthropic’s predominantly software‑driven stack. By stepping back, Anthropic may be signaling a strategic pivot toward organic growth, focusing its capital on compute rentals and talent acquisition rather than sprawling acquisitions.

Both developments point to a critical balancing act that AI companies must manage: securing massive compute resources while maintaining financial discipline. OpenAI’s approach leans heavily on securing physical infrastructure, effectively building a private “cloud” in a region where energy costs are competitive and regulatory environments are becoming more supportive of AI. Anthropic, on the other hand, appears to be tightening its belt, perhaps learning from the industry’s recent over‑investment cycles that have seen valuations soar, only to be corrected by market realities.

The involvement of Nvidia, indirectly referenced through Firmus’s backing, adds another layer of complexity. Nvidia’s GPU dominance gives it leverage not only in the hardware market but also in shaping the geography of AI compute. By supporting Firmus’s expansion, Nvidia is helping to decentralize AI workloads away from the traditional West‑coast data‑center clusters in the United States. This strategic dispersion could reduce latency for Asian users, lower operating costs, and provide a buffer against potential geopolitical restrictions on AI technology transfers.

For smaller AI start‑ups watching these moves, the lesson is clear: the future of AI development will be defined as much by where you can plug into massive power supplies as by the sophistication of your algorithms. Companies that can secure low‑cost, high‑capacity compute in emerging markets may gain a competitive edge, while those that overextend through costly acquisitions risk wasting capital that could otherwise fund compute rentals or research talent.

In the end, the AI battlefield is being redrawn on both the silicon and the geopolitical maps. OpenAI’s Malaysia partnership is a testament to a proactive, infrastructure‑first playbook, whereas Anthropic’s retreat from the Decart deal highlights the growing caution around oversized M&A in a sector still grappling with rapid cost escalations. As the industry moves forward, we can expect more of these strategic realignments, with compute availability and fiscal prudence rising to the forefront of AI development strategies.