Hardware & infra
Key Questions
What is the scale of hyperscaler capex in current AI infrastructure?
Hyperscaler capital expenditures are reaching a climax as demand for AI infrastructure surges. This includes major investments across chips, data centers, and supporting hardware.
What is AMD investing in Taiwan and why?
AMD is committing $10 billion in Taiwan to accelerate development of AI chips including TSMC 2nm processes and MI450X. The investment deepens partnerships for next-generation AI hardware production.
How is Dell redefining enterprise AI infrastructure?
Dell has introduced AI factories that rewire enterprise setups for large-scale AI workloads. These solutions integrate hardware and software optimized for production AI deployments.
What infrastructure bottleneck does IREN highlight for AI growth?
IREN's co-founder identifies power and infrastructure constraints, rather than chip availability, as the primary limit on AI scaling. The company notes a 5GW bottleneck affecting expansion.
What realities are emerging around Kubernetes for AI workloads?
Practical challenges in using Kubernetes for AI infrastructure include orchestration complexity and resource management at scale. Sessions at recent events address these production realities.
How is Nvidia extending NVLink technology?
Nvidia is advancing NVLink Fusion to improve connectivity and performance in AI systems. This supports tighter integration across accelerators in large-scale deployments.
What chip collaborations involve Microsoft and Anthropic?
Anthropic is in talks with Microsoft to run Claude models on custom AI chips. This effort aims to reduce reliance on Nvidia hardware while scaling inference.
What inference speeds is Cerebras achieving?
Cerebras has demonstrated 1200 tokens per second inference performance. This highlights ongoing hardware advances for high-throughput AI applications.
Hyperscaler capex climax; AMD $10B Taiwan (TSMC 2nm/MI450X); Dell AI factories; IREN 5GW bottleneck; Kubernetes AI realities. Ongoing: Nvidia NVLink Fusion, Microsoft-Anthropic chips, TPU DC scaling, Cerebras 1200 t/s inference.