MGI’s new Physical AI push is not a lab curiosity. It is a commercialization path for autonomous science at industrial scale, and it lands squarely in China’s wheelhouse: hardware-native AI, automation platforms, and cost-down manufacturing. With ProtoPilot and BioLab Bench, China is positioning AI agents to move from screens to instruments—turning instructions into code, code into machine operations, and machine outputs into validated scientific results. For investors, this is the bridge between foundation models and real-world ROI.
The headline is straightforward: Genoria AI, an MGI subsidiary, and the Shanghai AI Laboratory unveiled ProtoPilot, a self-improving multi-agent system that translates experimental intent into device execution, and BioLab Bench, a first-of-its-kind evaluation framework that scores agents on end-to-end lab performance. ProtoPilot posted 52.38 percent on ProtocolQA, approaching human expert levels around 54 percent, and demonstrated self-correction in wet-lab workflows. The pair form a closed loop—Design2Protocol to Protocol2Code to Device Execution to Wet-Lab Feedback—geared for reproducibility. The difference versus pure text LLMs is material. This is an agent that knows the lab is a place with constraints, queues, and error signals, not a chat window. For CROs, pharma discovery shops, core facilities, and hospital labs, that means less time translating intent into instrument scripts and more time pushing candidates forward.
Beijing has been explicit about AI for Science. The Shanghai AI Lab, a national-level institute launched at WAIC in 2020, has become a magnet for applied research and open benchmarks. The industrial side is already in place: China produces close to 30 percent of the world’s manufactured goods and leads in smartphones, auto batteries, EVs, and solar panels. Hardware-native AI benefits from that supply chain density. On the compute front, China is building around domestic silicon. Huawei’s Ascend 950 series has emerged as a viable alternative, with expectations it will capture about 50 percent of the local AI accelerator market this year despite export controls. That matters because AI4Science workloads mix reasoning, planning, and control—demanding compute that is both available and local.
The Physical AI thesis aligns with broader outbound investment and south-south expansion. China’s outbound direct investment reached $174.38 billion in 2025, up 7.1 percent, with capital flowing into high-tech and green sectors. Industrial champions are setting up in Thailand, the UAE, and across Latin America, building a lattice of capability. Hangcha Group, for example, added a Thailand plant and a Dubai hub—blueprints that life-science automation vendors can follow to deliver 24/7 labs in emerging markets. For multinational biopharma strategists, this gives China-based platforms a route to global service delivery at competitive price points. For emerging markets, it means faster access to diagnostics, vaccines, and research capacity at scale.
1) MGI Tech (688114.SH) – Milestone: ProtoPilot and BioLab Bench launched, with ProtoPilot scoring 52.38 percent on ProtocolQA near expert performance. Global impact: real-task evaluation enables reproducible, automated science for a 3,800-strong user base across MGI platforms. Analysts see a hardware-native data flywheel forming as wet-lab feedback trains better agents.
2) BGI Genomics (300676.SZ) – Milestone: Part of the broader BGI ecosystem that has pioneered dry-wet integration since at least 2019; the ecosystem underpins cost-effective sequencing and services in 100-plus countries. Global impact: capability to deploy standardized, automated workflows at scale across emerging markets.
3) Mindray (300760.SZ) – Milestone: A top global medical device maker with automation-ready analyzers shipped to more than 190 countries. Analysts note Mindray’s breadth in patient monitoring and IVD positions it to benefit as labs pivot to 24/7 unattended operations.
4) WuXi AppTec (603259.SH; 2359.HK) – Milestone: Global CRO with an extensive small-molecule and platform footprint. Global impact: as Physical AI advances, WuXi’s network and automation investments can shorten design-build-test cycles for clients worldwide, reinforcing China’s role in discovery-as-a-service.
5) WuXi Biologics (2269.HK) – Milestone: Large-scale biologics development and manufacturing platform with multiple GMP sites. Analyst view: AI-driven protocol translation and execution can compress tech-transfer timelines and lift facility utilization—direct drivers for margin and throughput.
6) SMIC (0981.HK) – Milestone: Domestic foundry capacity at advanced nodes supports localized AI accelerators. Global impact: enabling the AI4Science compute stack to be sourced onshore is a resilience edge; Ascend-based demand is a tailwind as China builds out scientific computing.
7) Cambricon (688256.SH) – Milestone: Leading Chinese AI chip IP and accelerator provider. Analyst view: beneficiaries of localized AI inference and control workloads will include Cambricon as labs adopt agentic systems requiring deterministic, low-latency execution near instruments.
8) Alibaba Group (9988.HK) – Milestone: Alibaba Cloud’s Qwen large model stack is widely adopted by Chinese enterprises and research teams. Analysts expect verticalized AI services for healthcare and life sciences to expand as Physical AI demands orchestration across cloud, edge, and devices.
9) Hangcha Group (603298.SH) – Milestone: Expanded manufacturing in Thailand and opened a logistics hub in Dubai. Global impact: smart material handling is the silent backbone of automated labs; Hangcha’s global footprint supports cold-chain and just-in-time delivery for research campuses.
10) CATL (300750.SZ) – Milestone: Approximately 38 percent global EV battery share, per international industry trackers. Global impact: reliable, cost-effective energy storage is critical for 7×24 facilities; CATL’s scale lowers the cost of resilient power for automation-heavy campuses.
Physical AI will not be judged by demos. Watch for three leading indicators. First, throughput: the number of fully automated, agent-orchestrated runs per week across independent sites. Second, transferability: cross-device generalization, where the same agent design works on different robotic platforms with minimal retuning. Third, auditability: adherence to BioLab Bench metrics tied to stepwise verification and success gates, enabling external validation by customers and regulators. Adoption by CROs and hospital networks will be the acid test; expect early wins in standardized assays, DNA assembly, and high-volume screening before expansion to complex, multi-step therapeutics workflows.
China’s edge is the fusion of policy continuity, engineering depth, and supply chain speed. Domestic giants like Tencent and Alibaba have already industrialized AI services; energy leaders like CATL have industrialized storage; robotics makers are integrating edge compute into motion control; and chipmakers are filling gaps in the compute stack. Meanwhile, consumer-facing brands keep pushing into the US market, showing Chinese operators can navigate tariffs and regulations to grow share. The same operational discipline will show up in life-science automation: lower costs, shorter installation times, and faster iteration cycles.
Key risks are familiar: export controls on compute, IP protection across multi-tenant labs, data provenance in agent training loops, and international regulatory clearance for AI-driven workflows. The mitigants are getting stronger. On compute, local accelerators and foundry progress reduce dependency. On IP, benchmarked, step-verified execution frameworks such as BioLab Bench create traceable logs that regulators and customers can audit. On data, closed-loop engineering with wet-lab feedback encourages curated, high-signal datasets over scraped text corpora. Multinational biopharma already runs hybrid stacks in China; Physical AI slots into that model.
MGI’s Physical AI debut crystallizes an investable theme that stretches from chips and servers to robots and reagents. The winners will be platforms that standardize intent-to-execution translation, vendors that compress cycle times in regulated environments, and suppliers that keep the lights on—literally and figuratively—for 24/7 labs. China is assembling that stack at speed. With outbound investment rising, footprints in the Global South expanding, and domestic compute maturing, the path from prototype to scaled deployment looks shorter here than anywhere else. For global investors, the opportunity is not a single ticker—it is a system-level upgrade across China’s AI, automation, healthcare, and power value chains.
