A recent launch event in Shanghai aimed to redefine the foundational logic of global scientific research, happening against a backdrop of parallel efforts in both China and the US to rewrite this very logic. Just a week prior, the US White House Office of Science and Technology Policy (OSTP) released a landmark report, "Science: A New Golden Age," reminiscent of the 1945 "Science: The Endless Frontier," elevating AI-driven automated closed-loop laboratories, scientific large models, and AI-native research organizations to the level of national strategy. The report explicitly states that future research competitiveness hinges on the ability to close the loop between "hypothesis generation, automated experimentation, and data feedback."
Coincidentally, every aspect outlined in the report from across the Pacific—an AI research operating system, automated closed-loop labs, and high-quality negative sample databases—has already been implemented by XTALPI (02228), and done so ahead of the curve. On July 29th, XTALPI (02228) officially launched the XtalPi Science platform and the Genius Agents matrix, while also spearheading the formation of an open science intelligence ecosystem alliance. This is not merely a product launch; it is a pre-emptive answer submitted by a Chinese tech company on the very track defined by a US national strategy report. This innovative platform company, driven by quantum physics, AI, and robotics, is completing a transformation from a "technology provider for AI for Science" to an "operator of scientific intelligence infrastructure." After over a decade of technological breakthroughs, its expanding "technology reservoir" is now rapidly productizing, opening a window for a valuation paradigm shift in the capital markets.
Bridging the Virtual-Physical Gap: Physical AI Captures the "Pricing Power" of AI's Next Phase
To understand the value of XtalPi Science, one must grasp the underlying logic of industry transformation. A popular saying in investment circles goes: "Computing power for large models has limits, but the physical world is infinite. In the next phase of AI, whoever can bridge the chasm between the virtual and the real will capture the pricing power of a trillion-dollar blue ocean." As large models become widespread, the marginal cost of generating scientific hypotheses with AI approaches zero. However, the scarcity of physical experimental validation has become the core bottleneck hindering the large-scale industrialization of AI for Science. Of the over $2 trillion spent globally on R&D annually, vast resources are wasted on blind trial-and-error and inefficient coordination of fragmented tools, two structural pain points that have long constrained industrial upgrading.
Most AI tools on the market stop at "screen-level conversations," where AI hallucinations can easily lead to losses of millions in real-world experiments and significant sunk opportunity costs. A deeper constraint comes from the "structural flaws" within data systems: public scientific databases only record "positive samples" from successful experiments, while failure data, which constitutes 80% of the real R&D process, is consistently lost. The tacit knowledge of experienced scientists is difficult to standardize and capture, and talent mobility directly leads to a loss of R&D capability, creating a natural gap between AI model predictions and real-world validation. Furthermore, tool fragmentation breeds data silos. In new drug and new material development, systems for each R&D stage operate independently with incompatible standards, forcing cross-stage collaboration to rely heavily on manual integration, passively lengthening development cycles. Standardized scientific workflows are difficult to replicate at scale, raising the cost of intelligent transformation for small and medium-sized tech entities.
Adding to the convergence, the US OSTP report "Science: A New Golden Age" precisely lists "building automated closed-loop laboratories" and "creating high-quality scientific databases" as core strategic initiatives. The report argues that the bottleneck for AI in science has shifted from algorithms and computing power to "physical interaction and data closure"—a direction perfectly aligned with XTALPI's strategic positioning over the past decade. As Wen Shuhao, Chairman of XTALPI, stated, the core of the next AI competition is no longer a numbers game of computing power consumption, but the ability to translate intelligence into real industrial growth. High-value R&D fields like life sciences and advanced materials are prime soil for Physical AI to achieve a commercial closed loop. Whoever can first establish the complete chain of "digital hypothesis → physical validation → data feedback" will gain a strategic advantage in capturing the trillion-dollar blue ocean market.
Pain Points Are Opportunities
In June of this year, China's National Data Administration explicitly designated real-machine interaction and embodied experiment datasets as key areas for development, with policy support increasingly favoring platforms with proprietary experimental data output capabilities. Both China and the US are pointing in the same direction at the same time, and XTALPI is already positioned ahead of the curve.
Full-Stack Closure + Proprietary Data: The US Plans It, XTALPI Has Built It
Facing common industry bottlenecks, XTALPI chose not to optimize single-point tools but instead built a complete closed-loop system spanning algorithms to experiments. The XtalPi Science platform is the standardized, productized embodiment of this capability. According to Zhitong Finance & Economics, XtalPi Science is the world's first AI4S-native operating system to deeply integrate large language models, a multi-agent matrix, and a large-scale automated robotics laboratory—making this "world's first" claim particularly significant. At a time when the US "New Golden Age" report has made it a national strategic direction, XTALPI's product is already operational.
XtalPi Science packages the models, tools, and experimental resources accumulated by XTALPI over ten years into standardized, callable capabilities, transitioning from an internal project tool to a globally accessible scientific research infrastructure. It reconstructs the traditional reliance on "luck and experience" in research into a systematic, orchestrated, verifiable, and traceable engineering process, advancing AI from merely "generating answers" to "driving verifiable scientific discovery."
The "brain" of this system is the Genius Agents matrix. It is not a single AI assistant but a complete virtual R&D organization capable of autonomously decomposing long-term, interdisciplinary research tasks, reconstructing the traditional DMTA R&D cycle, and compressing multi-week cycles into days. It covers the entire chain from literature review and molecular design to ADMET prediction and automated testing. It can solidify the tacit knowledge of scientists into standardized workflows while retaining the final decision-making authority of human experts, creating an efficient collaboration model of "AI-powered large-scale exploration + scientist-directed strategic direction." Ma Jian, CEO of XTALPI, noted that Genius Agents bridging the gap between large model reasoning and physical experiments is a key step in AI's evolution from a "chat tool" to an "engine of scientific discovery."
The true long-term moat, however, is built by the underlying Physical AI system and its proprietary data assets. While the US "New Golden Age" report calls for "establishing AI experimental validation and automated closed-loop laboratories," XTALPI's first-generation automated laboratory was already operational in 2020. Instead of traditional rule-driven automation, they built a self-learning cycle of "perception, decision, and execution." Currently, this system covers over 20 R&D scenarios, has served more than 100 new drug and new material projects, and boasts significant first-mover and scale advantages globally. The laboratory has accumulated over 500,000 complete experimental records, 80% of which are industry-scarce failure negative samples—precisely the core component of the "high-quality scientific databases" repeatedly emphasized in the "New Golden Age" report. The Sure Route retrosynthesis model, trained on this data, achieves a chemical hallucination rate of only 4.6%, just one-sixth that of leading general-purpose large models. Its failure prediction accuracy (81%-89%) far surpasses the benchmark of senior chemists (38%-60%). In commercial applications, the number of molecular synthesis iterations has been reduced from the traditional 5-10 rounds to an average of 1.19, directly cutting material and labor costs and achieving an order-of-magnitude improvement in R&D efficiency. This is not optimization; it is reconstruction.
Industry-Academia Alliance Validates Platform's Commercial Versatility
Beyond the technical closed loop, XTALPI is amplifying the platform's value and accelerating its commercial penetration by fostering an open ecosystem, which also validates the cross-track versatility of XtalPi Science. The concurrently launched Scientific Intelligence Open Ecosystem Alliance includes 27 initial members covering pharmaceutical companies, new material firms, instrument manufacturers, computing power providers, universities, and investment institutions. Using XtalPi Science as a unified base, the alliance offers diverse cooperation models such as open platform access, joint development, and private deployment, paired with a unified Science Token resource metering system. This model balances affordable access for smaller institutions with the data security needs of large enterprises, for whom it can build dedicated digital R&D systems. By the day of the launch, nearly 200 institutions had already submitted trial applications. The market's speed of "voting with its feet" is often more telling than analytical reports. The rapid expansion of the ecosystem validates two core logics: firstly, the versatility of XtalPi Science as an underlying infrastructure extends far beyond the single track of innovative drugs, capable of spilling over into fields like new materials and new energy; and secondly, the open ecosystem will further enrich data sources and strengthen customer stickiness, in turn reinforcing the platform's technological and data moats, creating a "wider ecosystem → more data → stronger moat" flywheel effect.
Business Model Evolution: From "Linear" to "Compounding" Valuation
From an investment perspective, the launch of XtalPi Science is poised to break through the growth bottleneck of the labor-intensive, project-based model, establishing a multi-layered, anti-cyclical growth framework and prompting the capital market to reassess the company's valuation logic. The company's business model has completed a structural shift from a "linear project-based approach" to a "platform-based compounding model." Revenue generation has moved beyond single project delivery, forming a diversified system encompassing platform subscriptions, private deployments, and integrated laboratory deliveries. Having expanded beyond innovative drugs, it is now replicating its mature material intelligence R&D capabilities into trillion-dollar new material sectors like lithium batteries, industrial catalysis, and fine chemicals, effectively smoothing out cyclicity in any single industry. The diversified revenue matrix exhibits compounding characteristics. As the platform continues to roll out, the company's revenue structure will increasingly shift towards high-margin, reusable subscription-based services, steadily improving overall profitability and growth certainty.
Even more noteworthy is the logic for revaluing its data assets. The "New Golden Age" report positions a "high-quality scientific database" as a national strategic resource. The 500,000+ real experimental records accumulated by XTALPI (80% exclusive failure negative samples), many comprising complete project and process data, effectively constitute the enterprise-level version of this strategic resource. This vast amount of proprietary, standardized experimental data allows XTALPI to add a core positioning of "scientific data asset operator," highly aligned with China's policy direction for capitalizing data elements. Against the backdrop of market-oriented data element reforms, these real-world R&D data can be monetized through platform-based usage fees, revenue sharing from joint development, and specialized data product offerings, creating a third growth curve alongside technical services and laboratory delivery. The high-precision, structured data collected by XTALPI's over 300 robots over the past five years, previously impossible to price independently, is now transforming from a byproduct of R&D into a core platform asset and a competitive barrier difficult for latecomers to match.
Operational data already validates this logic. In 2025, XTALPI achieved total revenue of 803 million RMB, a year-on-year increase of 201.2%, and a net profit of 135 million RMB, making it one of the few companies in the Hong Kong-listed AI for Science sector to achieve stable full-year profitability. The company's performance forecast for the first half of 2026 indicates that, excluding the high base effect from a large upfront payment for a pipeline license in the same period last year, its organic revenue for the period is expected to be no less than 250 million RMB, a year-on-year increase of over 65%. Revenue from AI science solutions is expected to reach 180 million RMB, with growth accelerating to over 120%, demonstrating the sustained growth momentum of its platform transformation.
Conclusion
In 1945, Vannevar Bush's "Science: The Endless Frontier" laid the groundwork for post-war US technological dominance. In 2026, the US "Science: A New Golden Age" declares that AI will rewrite the rules of scientific research. And everything planned in that report, XTALPI has already deployed and validated in real-world R&D scenarios. From being a "water seller" to a "wave maker," XTALPI's action is not just a product launch but a validation of a thesis: in the AI4S race favored by both China and the US, a Chinese company can lead the way. Leveraging China's engineering talent pool, large model ecosystem, advanced supply chain advantages, and massive industrial scale, it is well-positioned to maintain its lead and unlock the trillion-dollar blue ocean of scientific research. As traditional R&D models shift from "empirical trial-and-error" to "intelligent prediction," and as AI moves from the "chat window" into the "laboratory," XTALPI has used a decade of accumulated Physical AI and proprietary data assets to secure a precise position at the most critical juncture for AI for Science industrialization. It is poised to capitalize on the intelligent upgrade opportunities in biomedicine, new energy, and new materials, solidifying its leading position in the domestic AI4S field and achieving a dual enhancement of industrial and investment value. For the capital market, the window for cognitive change often coincides with the golden period for value reassessment.
Comments