According to a research report released by DFZQ (HKEX: 03958, SSE: 600958), general-purpose large models are also expected to capture market opportunities in the embodied intelligence sector. The firm believes that the entry of major model developers will accelerate progress in embodied models, and therefore expectations for the industry's commercialization are continuously rising. The firm notes that major model developers like OpenAI are expanding into the embodied intelligence space, which is expected to drive breakthroughs in more application scenarios, thereby accelerating the industry's commercialization process, with robot complete-machine and component manufacturers positioned to benefit.
Expansion of Large Model Robot Applications Drives Higher Commercialization Expectations
The firm observes that the previously pessimistic market sentiment toward the embodied intelligence industry was mainly due to the belief that embodied models differ from large models, along with concerns that current embodied industry players are making slow progress on models, resulting in limited application scenarios for embodied intelligence. However, the firm notes that GPT-6 Astra has already achieved solid application practices in the embodied field, and some investors are gradually coming to believe that general-purpose large models can also capture market opportunities in the embodied sector. The firm believes that the entry of major model developers will accelerate progress in embodied models, and therefore the industry's commercialization expectations are continuously rising.
Large Models Show Huge Potential as Embodied Strategy Layers, Embodied Body Manufacturers Benefit Most at the Margin
Galbot previously tested Astra and π0.5 in a dual-arm RoboDojo subset containing ten tasks. The results showed that using Astra alone yielded only a 26% success rate, while mixed control combining Astra and π0.5 raised the task success rate to 48%. The firm believes these experimental results partially reveal the potential of GPT-6 Astra as a strategy-layer brain 鈥?that is, using Astra to understand tasks and thereby calling more basic embodied models to accomplish task objectives. For embodied body manufacturers, their trained underlying models can be combined more efficiently, further enhancing deployment capabilities in actual implementation. The firm believes that as general-purpose large model developers continue to explore the embodied intelligence field, this is expected to drive accelerated application of embodied intelligence in logistics and industrial manufacturing, with manufacturers possessing vertical-scenario know-how enjoying relatively higher barriers to entry.
Major Model Developers Accelerating Embodied Intelligence Layout Set to Boost Industry Commercialization Expectations
The firm notes that many major model developers have recently been continuously expanding their embodied intelligence layouts, which is expected to catalyze higher commercialization expectations for the industry. OpenAI recently announced it will produce impressive demonstration robot prototypes by 2027, Tesla is expanding its Optimus robot data-collection team, and among domestic players, ByteDance is also expanding its robotics engineer team. The firm observes that as these major model developers accelerate their embodied intelligence layouts, it expects more large model developers to follow suit. The firm believes this is expected to drive breakthroughs in more application scenarios, thereby accelerating the industry's commercialization process, with robot complete-machine and component manufacturers positioned to benefit.
Risk Warnings
Risks include model development and data accumulation progressing slower than expected, unclear scenario demand implementation, manufacturer production falling short of expectations, changes in national policies, and industry financing falling below expectations.