Three layers of AI chips explain Nvidia, Broadcom, Intel reactions

Core insights:

• The AI chip sector consists of three distinct levels: mass-market chips, tailored chips and fabrication.
• General-purpose chipmakers like Nvidia gain most from widespread AI spending.
• Tailored chip designers such as Broadcom benefit when large AI firms look for alternatives, though they face customer concentration and financial risks.
• Foundries need big committed buyers to cover the cost of expensive plants, which is why customer announcements affect Intel's shares so strongly.
• Electricity supply, supplier diversity and manufacturing centralisation are influencing all levels.

A guide to the AI chip race covering GPUs, ASICs and foundries.

What explains a single AI headline sending Nvidia to a new high while Intel falls on the same day? The reason is that "AI chips" covers several businesses. Companies operate in different layers of the industry, each with unique economics, risks and success factors. Grasping these layers makes it far simpler to follow sector news and understand why investors respond so differently to it.

Important terminology:

• GPU (graphics processing unit): A general-purpose processor capable of handling many AI workloads. Nvidia's GPUs set the standard for training and deploying large AI models.
• ASIC (application-specific integrated circuit): A chip built for a single customer or purpose. It lacks the flexibility of a GPU but can be cheaper or more power-efficient for its intended use.
• TPU (tensor processing unit): Google's custom AI chip, an ASIC example, co-designed with Broadcom.
• Foundry: A manufacturer that produces chips designed by other firms. TSMC is the biggest foundry.
• Anchor customer: A large, committed buyer that takes up most of a new factory's capacity, helping to justify the investment.
• Process node: A generation of chip-making technology. Each new node produces more capable chips but requires huge upfront spending.

The first layer: chips for general AI use

This layer covers the chips that train and run AI systems. Nvidia leads here. Its edge comes from more than just hardware: software developers have long built on its platform, making it costly to switch.

Firms in this layer do well when AI spending is strong and across many customers, because their chips can serve almost anyone. In early October 2026, Nvidia hit a record high as investors focused on robust revenue forecasts and a big share buyback.

Second layer: chips built for a single client

This layer involves custom chips designed for one major buyer. The biggest AI developers spend so much on computing that it becomes worthwhile to tailor hardware to their own models, potentially lowering costs and reducing reliance on a single supplier.

Broadcom is a leading designer in this space. Committing to custom chips is expensive, so new financing models are appearing. One example is leasing: a separate investment vehicle buys the chips and rents them to the AI company, spreading the cost over time and bringing in lenders and investors. In early October 2026, Broadcom was reported to be arranging $60 billion in such financing for custom chips used by Anthropic.

The main risks in this layer are concentration, because each design depends heavily on one customer, and the complexity of the financing that supports it.

Third layer: chip production

This layer is about manufacturing the chips. No matter who designs a chip, it must usually be produced by a foundry. Advanced factories cost tens of billions of dollars, so foundries need reliable, high-volume customers to fill them and pay for the next process node.

TSMC dominates this layer. Intel is trying to build a competing foundry business, which is why customer wins are critical to its stock price. In early October 2026, Intel shares dropped after Elon Musk confirmed that TSMC was in talks to join Terafab, a Texas chip project where Intel had been the only named manufacturing partner. The worry was not a lost contract but a threat to a customer win that investors had already priced in.

Interpreting chip news based on layers

The layered perspective clarifies otherwise puzzling market moves.

News of strong AI demand usually boosts layer-one firms most, as they sell to a broad base. News that a big AI developer is making its own chips can help layer-two designers while raising doubts for layer-one suppliers. News about factory partnerships mainly affects layer-three companies, and matters most for those that are still building their reputation.

Valuations add another factor. Companies with visible, current demand can withstand bad headlines more easily. Firms whose share prices depend on future wins are more sensitive to anything that makes those wins less certain.

Cross-layer forces

Several long-term trends affect the whole industry. Power supply is becoming a bottleneck, so chips that deliver more computing per watt gain an advantage. Large buyers want alternatives to any single supplier, which encourages custom chip design. And advanced manufacturing remains concentrated in very few places, giving the leading foundry strong bargaining power.

Tracking how these trends evolve is often more valuable than reacting to any single headline.