The AI Race: Who Is Building Real Technology, and Who Is Just Wrapping It

Thousands of companies call themselves AI companies. Only a small number build the foundations everyone else relies on. Understanding the layers of the AI stack helps cut through the hype.

Layer 1: Chips and manufacturing

Nvidia dominates AI accelerators, with AMD, Google’s TPUs and custom chips from Amazon, Microsoft and Meta competing. TSMC manufactures most advanced chips, and the Netherlands’ ASML makes the lithography machines that make them possible. China is building alternatives, led by Huawei.

Layer 2: Compute and data centres

Microsoft, Amazon, Google, Oracle and Meta are spending enormous sums on data centres and power. In China, Alibaba, Tencent, ByteDance and Baidu lead. Europe is funding AI factories and supercomputers.

Layer 3: Frontier models

• United States: OpenAI, Anthropic, Google DeepMind, Meta and xAI.
• China: DeepSeek, Alibaba’s Qwen team, Moonshot AI, Zhipu AI, MiniMax, ByteDance and Baidu.
• Europe: Mistral AI, plus specialised firms such as DeepL and Black Forest Labs.

Layer 4: Applications

This is where most startups live: tools for writing, coding, customer support, legal work and more. Many add real value with proprietary data, deep workflow integration or specialised fine tuning.

How to spot a thin wrapper

• No proprietary data, workflow or distribution advantage.
• The product is mostly a prompt and a nice interface on someone else’s model.
• A single model update from a big lab could make it obsolete.

Who is winning?

The US leads in frontier models, chips and capital. China is close behind on models and leads in open weight releases, despite chip restrictions. Europe leads in regulation and some specialised areas. The race is long, and energy, talent and chips will decide much of it.

Sources

• Stanford HAI, AI Index Report
• Epoch AI, data on AI models and compute