Custom AI chips represent one of the most significant threats to Nvidia's market dominance. Chief Executive Jensen Huang's response: find a way to profit from them anyway. Nvidia just injected $3.5 billion into a Taiwanese chip manufacturer to help customers build custom AI processors that can integrate into Nvidia's full system stack. Meanwhile, Anthropic has committed $35 billion to Lambda, a specialized AI cloud provider, for computing resources. These figures aren't directly comparable — the former is an investment amount, while the latter is a consumption commitment — but both transactions point to the same long-term strategy. Huang wants Nvidia to become the foundational architecture of AI, not merely a dominant chip supplier.

In simple terms, an AI factory takes in electricity and data and outputs tokens — the "currency" of AI. As Huang puts it: "Electrons go in, tokens come out." During the most recent earnings call, Nvidia indicated that the market value of a single AI factory's computing power has climbed from roughly $18 billion in the Hopper architecture era to $25 billion with Blackwell, and the Vera Rubin platform has already reached $40 billion. Vera Rubin isn't just a new GPU; the platform combines Nvidia's latest CPU and GPU with proprietary networking, memory, and other supporting infrastructure. This incremental revenue comes from selling the entire AI factory ecosystem around the GPU.

MediaTek has pushed this strategy to a new level. Clients can develop their own chips that compete with Nvidia GPUs, then use NVLink Fusion technology to connect those chips into Nvidia's complete data center architecture. Nvidia hasn't disclosed how much revenue an AI factory generates when a third party supplies the primary AI chip. But losing the chip order no longer means losing the entire business. Amazon Web Services (AWS) serves as a real-world example of this model. AWS plans to deploy two million Nvidia GPUs while also deeply integrating its proprietary Trainium chips into Nvidia's architecture. Whichever path customers choose, Nvidia still captures a share of the value.

Key moves in Nvidia's AI factory expansion

Date: 2025-05 — Initiative: Launched NVLink Fusion for custom chips — Strategic direction: Expanding foundational architecture ecosystem

Date: 2026-01 — Initiative: Invested $2 billion in CoreWeave — Strategic direction: Expanding computing capacity

Date: 2026-03 — Initiative: Invested $2 billion in Marvell — Strategic direction: Expanding foundational architecture ecosystem

Date: 2026-08 — Initiative: Wall Street financing plan targeting over $500 billion — Strategic direction: Expanding computing capacity

Date: 2026-08 — Initiative: AWS adds two million Nvidia GPUs while connecting Trainium to Nvidia's architecture — Strategic direction: Dual-track positioning

Date: 2026-08-31 — Initiative: Invested $3.5 billion in MediaTek — Strategic direction: Expanding foundational architecture ecosystem

Date: 2026-09-01 — Initiative: Anthropic commits $35 billion to Lambda for computing capacity — Strategic direction: Expanding computing capacity

Expanding computing capacity addresses another well-known AI bottleneck. The constraint used to be chip shortages; now capital has become a limiting factor as well. The value of an AI factory comes from the economic output it generates over many years. Nvidia is pushing Wall Street to view computing power the same way: as infrastructure that can be financed against future earnings. The company is advancing a plan aimed at unlocking more than $500 billion in capital.

By selling the core hardware of the AI boom, Nvidia has dramatically outperformed the broader chip sector. Now Huang's ambition extends beyond chips themselves to the entire AI infrastructure stack built on top of them. Over the past five years, Nvidia shares have climbed more than 870%, compared with a 230% gain in the benchmark chip index. The red dashed line marks the start of the current U.S. bull market in October 2022.