California water lawsuit, AMD-Cerebras deal, and SK Hynix: Data center infrastructure in flux

A 330 megawatt AI data center project in California has filed a lawsuit against the state's most powerful irrigation district after being denied access to Colorado River water for cooling. The facility, which would be the largest in the state, claims it needs 287 million gallons of water annually.

The case highlights the growing tension between hyperscale data center expansion and water scarcity in the arid West. The project originally sought to purchase a farm, shut off its irrigation, and redirect the water to cooling towers. The irrigation district refused, prompting the lawsuit.

AMD and Cerebras target inference efficiency

On the hardware side, AMD and Cerebras Systems announced a partnership targeting disaggregated inference. They claim a 5x improvement in tokens per watt compared to conventional GPU clusters. The approach mirrors Nvidia's $20 billion investment in SRAM decode technology, but AMD opted for a partnership instead of an acquisition.

• AMD's rack-scale AI system roadmaps aim to compete with Nvidia's dominance in the data center.
• The partnership focuses on inference workloads, which are growing faster than training as AI models are deployed.
• Cerebras's wafer-scale engine is paired with AMD's Instinct GPUs for a disaggregated architecture that separates compute from memory.
• The companies claim higher efficiency by optimizing data movement, a key bottleneck in large-scale inference.

Memory price stability becomes a Big Tech demand

SK Hynix, one of the world's largest memory manufacturers, reported that major cloud providers are demanding long-term pricing agreements that smooth out volatility. The company's executives said during an earnings call that Big Tech wants predictable costs for HBM and DDR5 memory to support AI infrastructure buildouts.

The confluence of water disputes, new chip architectures, and memory pricing pressures suggests that the AI infrastructure boom is entering a phase of resource contention and supply chain maturation. Data center operators will need to navigate both physical resource constraints and component market dynamics to sustain growth. The coming months will test whether regulatory frameworks and hardware innovations can keep pace with demand.