In August 2025, Beijing-based robotics firm Galbot announced that its G1 Premium humanoid robot is now running on Nvidia’s Jetson Thor platform. According to the source material, Galbot is the first company to integrate this particular Nvidia module into a humanoid robot. The announcement positions the G1 Premium as an early adopter of the Jetson Thor, a physical AI platform that Nvidia has been rolling out to select robotics developers.
The G1 Premium was demonstrated at the World Robotics Conference in Beijing earlier that month. The source material does not specify the exact dates of the conference, only that it took place in August 2025 and that Galbot showcased the robot there. The integration itself appears to have been completed in time for that demonstration, though the precise timeline of when Galbot received the hardware and when the integration was finished is not disclosed in the source material.
Galbot’s founder and CTO, Professor Wang He, is quoted in the source material as saying that the G1 Premium, now running on Nvidia Jetson Thor, has demonstrated “remarkable advancements in speed and improved real-time reasoning capability.” He also noted that the early adoption of the platform allows Galbot to push its proprietary VLA (vision-language-action) models to new levels of real-world capability. The source material does not provide additional details on the architecture of these VLA models, nor does it specify how they are trained or deployed.
The hardware upgrade itself is significant. According to Nvidia, the Jetson Thor module offers more than seven times the AI computing capacity of its predecessor, the Jetson Orion. The source material also states that the Thor provides more than three times the energy efficiency compared to the earlier generation. In a separate passage, the source material cites figures of 7.5 times the AI compute of the previous Nvidia Jetson Orin and 3.5 times higher energy efficiency. These numbers appear in different parts of the source material, and the discrepancy between them — seven versus 7.5, three versus 3.5 — is not reconciled. It is possible that one set of figures refers to a different comparison baseline or a different configuration, but the source material does not clarify this. What can be stated with confidence is that Nvidia claims a substantial generational leap in both compute capacity and energy efficiency for the Jetson Thor over its predecessor.
The G1 Premium is designed for deployment across three sectors: retail, healthcare, and logistics. The source material does not provide details on specific deployments, customer names, or pilot programs in any of these verticals. It also does not specify the robot’s physical specifications, payload capacity, or operating environment constraints. What is known is that Galbot has positioned the G1 as a general-purpose humanoid robot, and the company’s broader roadmap includes work on dexterous manipulation. A related reference in the source material points to a 2024 Nvidia technical blog post about Galbot building a large-scale dexterous hand dataset for humanoid robots using Nvidia Isaac Sim. That reference suggests Galbot has been working with Nvidia’s simulation tools for some time, though the source material does not elaborate on how the Isaac Sim work relates to the Jetson Thor integration.
The source material also notes that Galbot is among the top humanoid robot companies taking a different design philosophy from the rest of the field. The exact nature of that philosophy is not fully described in the source material, but the implication is that Galbot is not simply chasing a particular form factor or a narrow set of tasks. Instead, the company appears to be building toward general-purpose autonomy, with the Jetson Thor integration serving as a compute foundation for that goal.
Why it matters for European robot service
For European operators and integrators, the Galbot announcement carries several implications, even though the company is based in China and the source material does not mention any European deployments.
First, the integration of Nvidia’s Jetson Thor into a humanoid robot signals that the compute platform is maturing. Nvidia has positioned itself as a major supplier of tools for robotics development, and the Jetson modules are designed to be embedded in AI robots, combining Blackwell GPUs, the Isaac development platform, and sensor signal processing capabilities. The source material notes that most major Chinese robotics players are working with Nvidia products in some way, and that UBTech, Galbot, Unitree, EngineAI, and AgiBot were among the first to receive the latest Jetson modules. This means that the compute stack powering humanoid robots is increasingly standardised around Nvidia hardware, which has implications for European buyers who may be evaluating robots from multiple vendors.
If a European logistics operator is considering humanoid robots for warehouse tasks, the fact that multiple vendors are running on the same Nvidia compute platform could simplify certain aspects of evaluation. Software tools, simulation environments, and possibly even some middleware may be shared across vendors. However, the source material does not provide evidence of any such standardisation in practice. It only notes that Nvidia provides key tools for development work and that Chinese players are using them.
Second, the energy efficiency gains are relevant for European operators who are increasingly focused on sustainability and total cost of ownership. The source material states that the Jetson Thor offers more than three times the energy efficiency of its predecessor. For a robot that runs continuously in a retail or logistics environment, energy consumption is a meaningful operational cost. If Galbot’s G1 Premium can deliver the same or better performance while drawing less power, that could make the robot more attractive for European deployments. However, the source material does not provide absolute power consumption figures, so it is not possible to calculate actual energy costs or savings.
Third, the emphasis on real-time reasoning and complex planning is directly relevant to European service robotics use cases. Retail, healthcare, and logistics all involve unstructured environments where a robot must react to changing conditions. The source material states that Galbot’s robots can now perform complex planning and motion tasks with new levels of precision and efficiency, thanks to the Jetson Thor integration. For European buyers, this suggests that the G1 Premium may be capable of handling tasks that require more than simple pick-and-place operations. But again, the source material does not provide specific examples of tasks, benchmarks, or performance metrics beyond the general claims of improved speed and reasoning.
Fourth, the source material includes a reference to 1X, a Norwegian robotics company, and its NEO home robot. The NEO is priced at $20,000 for early access, targets delivery in 2026, and uses Nvidia’s Jetson Thor processor along with 1X’s proprietary Redwood VLA and World Model AI system. All inference runs on-device for safety-critical functions. 1X has also secured a deal with Swedish investment firm EQT to deploy up to 10,000 NEO units across EQT’s 300-plus portfolio companies. This is a separate development from Galbot’s announcement, but it is mentioned in the source material and is relevant to the European market because 1X is a European company and EQT is a European firm. The fact that two different humanoid robot companies — one Chinese, one Norwegian — are both building on Nvidia’s Jetson Thor suggests that this compute platform is becoming a common foundation for the industry. For European buyers, this could mean that the software ecosystem around Jetson Thor will grow, potentially leading to better support, more third-party tools, and more experienced integrators.
Fifth, the source material mentions that Nvidia’s Jensen Huang has been betting on Chinese robotics companies, giving the first batch of Jetson Thor chips to a Chinese recipient. The source material does not name that recipient, but it does note that several Chinese companies were among the first to receive the latest Jetson modules. This is relevant for European observers because it indicates that the most advanced compute hardware is flowing to Chinese firms first, which could create a temporary competitive advantage for those firms in terms of development speed. European robotics companies may need to consider whether they have access to the same hardware and when they can expect to integrate it into their own products.
Finally, the source material does not mention any European regulatory considerations, safety certifications, or data protection issues related to the G1 Premium. European buyers should be aware that deploying a Chinese-built humanoid robot in the EU may raise questions about data residency, cybersecurity, and compliance with local regulations. The source material does not address any of these topics, so they remain open questions.
What buyers and operators should know
For buyers and operators evaluating humanoid robots for retail, healthcare, or logistics, the Galbot G1 Premium with Nvidia Jetson Thor is worth watching, but there are several important caveats.
First, the performance claims are vendor-provided. The source material quotes Galbot’s founder and CTO, Professor Wang He, describing improvements in speed and real-time reasoning, and it cites Nvidia’s figures for compute capacity and energy efficiency. These are not independent benchmarks. Buyers should ask for specific performance data, ideally measured in their own environments or in standardised tests, before making procurement decisions. The source material does not provide any such data.
Second, the G1 Premium is designed for retail, healthcare, and logistics, but the source material does not specify which tasks it can perform in each sector. It does not state whether the robot can handle shelf stocking, patient assistance, parcel sorting, or any other specific function. Buyers should not assume that the G1 Premium is ready for a particular use case without direct evidence. The source material only says that the robot can perform “complex planning and motion tasks” with greater precision and efficiency, which is a general claim.
Third, pricing and availability are not disclosed in the source material. Unlike the 1X NEO, which has a stated price of $20,000 for early access and a $499 per month subscription model, the Galbot G1 Premium has no listed price, delivery timeline, or commercial terms. Buyers who are interested in the G1 Premium will need to contact Galbot directly for commercial information. The source material does not provide any contact details or ordering information.
Fourth, the source material does not mention any European distribution channels, service partners, or support infrastructure for Galbot robots. For European operators, this is a significant consideration. A robot is not a one-time purchase; it requires ongoing maintenance, software updates, spare parts, and potentially on-site support. The source material does not disclose whether Galbot has any presence in Europe, whether it works with local integrators, or what its service level commitments are. Buyers should not assume that support will be available locally.
Fifth, the source material does not provide any information about safety certifications, standards compliance, or liability frameworks for the G1 Premium. Humanoid robots operating in retail, healthcare, or logistics environments will need to meet local safety requirements, and it is not clear from the source material whether the G1 Premium has been certified for any market. Buyers should ask for documentation on certifications and standards compliance before committing to a deployment.
Sixth, the source material mentions that Galbot is among the top humanoid robot companies taking a different design philosophy from the rest of the field. The exact nature of that philosophy is not described in detail, but it may be relevant to buyers who are comparing different humanoid robots. Some companies focus on bipedal locomotion, others on dexterous manipulation, and others on specific vertical applications. Galbot’s approach appears to be general-purpose, with an emphasis on VLA models and real-world capability. Buyers should understand what this means in practice, and the source material does not provide enough detail to fully characterise Galbot’s design philosophy.
Seventh, the source material does not disclose the timeline for when the G1 Premium will be commercially available, if it is not already. The robot was demonstrated at the World Robotics Conference in August 2025, but demonstration does not equal commercial availability. Buyers should ask Galbot for a clear product roadmap, including availability dates, production volumes, and any early access programs.
Eighth, the source material does not provide any information about the total cost of ownership for the G1 Premium. Beyond the purchase price, buyers will need to consider energy costs, maintenance, software licensing, and potential downtime. The energy efficiency improvements from the Jetson Thor are a positive sign, but without absolute power consumption figures, it is not possible to estimate annual energy costs.
Ninth, the source material does not mention any warranty, service level agreements, or response time commitments from Galbot. Buyers should not assume that any such commitments exist. The source material explicitly does not provide SLA numbers, response times, or spare-part lead times, and any such figures would be invented if stated here.
Tenth, buyers should consider the broader ecosystem. The source material notes that Nvidia’s Jetson modules combine Blackwell GPUs, the Isaac development platform, and sensor signal processing capabilities. This means that software developed for one Jetson-based robot may be partially portable to another, which could reduce switching costs for buyers who standardise on Nvidia-based platforms. However, the source material does not provide evidence of such portability in practice.
In summary, the Galbot G1 Premium with Nvidia Jetson Thor represents a notable technical milestone — the first integration of this compute platform into a humanoid robot — and the performance claims are significant. But for European buyers, the lack of disclosed commercial terms, support infrastructure, safety certifications, and independent performance data means that a purchase decision should be made with caution. The source material provides a snapshot of an announcement, not a comprehensive product evaluation. Buyers should seek additional information directly from Galbot and should consider running their own pilots before committing to deployment.
Sources
Galbot becomes first company ‘in the world’ to integrate Nvidia Jetson Thor into a humanoid robot
Published by Vigla Media OÜ (Estonia).