NVIDIA raises Taiwan spend to $150B/year and begins first Vera CPU deliveries

Jensen Huang announced on May 28 that NVIDIA is increasing its annual Taiwan manufacturing and supply-chain spend from $100 billion to $150 billion to meet what he characterized as an inference-driven AI demand inflection. The first Vera CPUs — NVIDIA's Arm-based data center CPU paired with its Rubin GPUs — have shipped to leading AI labs and clouds, with broader availability expected in H2 2026. Huang floated a path to $3 trillion in cumulative revenue driven by agentic inference workloads, an order-of-magnitude framing meant to signal how big NVIDIA thinks the inference compute market will become.

The NVIDIA AI official X account on May 29 teased 'A new era of PC' from Taipei coordinates — widely read as a Computex 2026 reveal of a consumer-class Vera or DGX-Spark-style desktop AI machine, though NVIDIA hasn't confirmed specifics. The post landed alongside the Vera CPU shipping news, suggesting NVIDIA is pushing its inference stack down into prosumer and developer workstations, not just hyperscaler data centers.

Competitive context: the $150B Taiwan commitment is partly defensive against geopolitical risk (TSMC concentration) and partly offensive (locking up advanced packaging capacity). Vera CPU shipping means NVIDIA now offers a fully integrated CPU+GPU stack that competes more directly with AMD's MI400 and the cloud-provider custom silicon (AWS Trainium, Google TPU, Microsoft Maia). NVIDIA also distributed an NVFP4-quantized DeepSeek V4-Pro on Hugging Face this week — confirming its 'optimize and host every important model' partner strategy regardless of geopolitical noise.

Watch next: Vera CPU benchmarks versus AWS Graviton and Google Axion, whether the $3T inference revenue framing survives DeepSeek-style price cuts, and what NVIDIA actually announces at Computex.