At the 2026 World Robot Congress, the founder and CEO of Star Map, Gao Jiyang, shared his perspective on the evolution of embodied intelligence industrialization. He believes that as foundation models continue to improve, the industry's business models will undergo several major transitions, with the core of industrial value gradually moving from complete hardware to intelligent capabilities, while hardware profit margins are expected to keep shrinking.

In the first phase, which ran before 2026, sales of complete robots dominated the market. During this period, foundation models had limited capabilities, making it difficult to achieve model-driven commercial deployment. Market transactions relied primarily on robot hardware sales, with gross margins for complete units holding steady between 40% and 60%, and competition centered on hardware products.

The second phase, spanning from the second half of 2026 through 2028, will see the rise of solution subscriptions. With rapid iterations in foundation models, embodied intelligence will develop complete industrial solutions for specific productivity scenarios, shifting the business model toward subscription-based offerings. Customers will mainly pay for the industrial solution itself, with robot hardware serving merely as the physical carrier for implementation. Industry forecasts suggest hardware gross margins will fall to around 20%, while the value of software and solutions becomes increasingly prominent.

The third phase will introduce a token sales model for the physical world. As intelligent versatility improves and robots break free from single-scenario constraints to operate across multiple domains, the embodied intelligence sector will experience a second business model shift, centered on selling physical world tokens. To expand intelligent consumption across the physical world, massive numbers of terminal robots will need to be deployed, and at that stage, hardware may even enter a negative gross margin state, with all industry profits derived from intelligent services.

According to Gao, the embodied intelligence industry can be divided into two segments: complete hardware and intelligent capabilities, with value steadily migrating from the hardware side to the intelligence side. Foundation models are the core force driving this transformation, and the explosive growth of global open-source foundation models over the past year, coupled with accelerating technological iteration, is pushing the industry faster toward the latter two commercialization stages.