XtalPi’s industrial AI reaches Fangda Carbon factory floor as graphite costs stay volatile

XtalPi Holdings Limited (HKEX: 2228) has moved one of its artificial intelligence systems out of the laboratory and into Fangda Carbon New Material Co., Ltd.’s (SSE: 600516) live graphite electrode production workflow, marking a potentially important step in XtalPi’s attempt to build an industrial AI business beyond pharmaceuticals. The jointly developed model evaluates raw-material combinations and blend ratios before physical production trials, allowing Fangda Carbon to narrow the number of formulations that need to be tested on the factory floor. The companies say the system has completed acceptance testing, achieved required predictive and cost-optimization targets and is already being used in Fangda Carbon’s raw-material selection process. The commercial question is now whether the model can produce measurable savings and repeatable manufacturing benefits at scale, because neither company has disclosed the percentage cost reduction, annual savings or financial contribution expected from the deployment.

The milestone is the first accepted core module in a broader graphite electrode formulation optimization program developed under the companies’ artificial intelligence collaboration. XtalPi and Fangda Carbon originally signed a strategic cooperation agreement in March 2025 covering intelligent research, production optimization and high-performance carbon materials, before the graphite electrode formulation project moved into more focused development work. The current deployment therefore represents something more concrete than another industrial AI pilot: the model has passed project acceptance and entered an operating production workflow, although final formulation decisions still require expert review and experimental or production validation.

Why does putting XtalPi’s AI inside Fangda Carbon production matter more than another industrial AI pilot?

Industrial artificial intelligence frequently attracts attention at the proof-of-concept stage, but commercial value only begins to become visible when a model survives validation and enters a workflow that influences real procurement or production decisions. Fangda Carbon’s graphite electrode model has crossed that threshold by moving into operational use after being tested against independent datasets and production trials. XtalPi says the system can rank candidate formulations quickly enough to match the pace required by an industrial manufacturing process.

The model addresses a practical problem in carbon manufacturing. Graphite electrode quality depends on interactions among raw-material properties, formulation ratios and production conditions, while supposedly similar feedstocks can behave differently when suppliers or production batches change. Manufacturers therefore face a recurring need to adjust recipes without compromising electrical conductivity, strength and other performance requirements.

Traditionally, a large part of that work relies on engineering experience combined with physical trials. The number of possible combinations can become costly and time-consuming when several raw materials, proportions and operating parameters change simultaneously. XtalPi’s model attempts to reduce that search space by predicting which combinations are most likely to satisfy required performance specifications before Fangda Carbon commits material and production resources to physical validation.

The important boundary is that the AI is not autonomously determining the finished commercial recipe. XtalPi says unsuitable combinations can be screened out and promising candidates ranked, but human specialists remain responsible for final formulation decisions and necessary experimental or production verification. The system therefore functions as a decision accelerator rather than a replacement for Fangda Carbon’s materials expertise.

How could predictive raw-material selection reduce graphite electrode manufacturing costs?

Graphite electrode economics are heavily influenced by raw materials, particularly petroleum and needle coke, together with energy-intensive processing. Prices and physical characteristics can vary across suppliers and batches, creating a difficult trade-off between sourcing cost and manufacturing performance. A cheaper input may offer little benefit if it produces lower-quality electrodes, requires process changes or increases failure rates.

The XtalPi and Fangda Carbon system approaches the problem by identifying combinations designed to meet specified quality requirements while minimizing raw-material costs. When prices or supply availability change, the model can evaluate substitutes and recommend revised proportions instead of forcing engineers to begin a broad physical testing program from the start.

That capability could become especially useful during periods of raw-material volatility or supply disruption. A manufacturer able to qualify alternative inputs more quickly gains additional procurement flexibility, potentially reducing dependence on a narrow group of suppliers. The economic benefit would come not only from selecting less expensive ingredients but from shortening the time and material consumed during formulation trials.

The companies have not quantified those economics. There is no disclosed percentage reduction in raw-material expense, number of physical trials eliminated, improvement in manufacturing yield or payback period for the AI project. Claims that the deployment will materially lift Fangda Carbon’s earnings would therefore be premature. The next level of evidence would be operating data showing how model-assisted formulations compare with the previous development process in cost, speed, quality consistency and production yield.

Why is Fangda Carbon an unusually useful test case for XtalPi’s industrial AI strategy?

Fangda Carbon brings decades of manufacturing records and specialist knowledge to a problem where proprietary data can matter as much as algorithmic sophistication. The companies say the development process involved organizing dispersed historical production data, building predictive features and combining performance forecasting with optimization algorithms and expert rules. That gives XtalPi access to the kind of real-world industrial data required to determine whether scientific AI models can work under factory constraints rather than controlled research conditions.

The underlying company also provides meaningful scale for validation. Fangda Carbon is a major Chinese producer of graphite electrodes and other carbon products, supplying materials used in metallurgy, new energy and industrial applications. Its main products include graphite electrodes, carbon bricks and other advanced carbon materials, giving the partnership potential expansion routes beyond the initial raw-material selection module.

Fangda Carbon’s latest financial results also show why operating efficiency has commercial relevance. For the first six months of 2026, the company reported revenue of approximately RMB1.97 billion, up 16.6% from RMB1.69 billion a year earlier. Net profit attributable to shareholders rose 213.5% to RMB170.9 million, although profit excluding non-recurring items increased a much more moderate 21% to RMB10.1 million, indicating that the headline profit increase was substantially influenced by items outside recurring operating earnings.

Reported cost of sales reached approximately RMB1.73 billion against RMB1.97 billion of revenue. Based on those figures, a Business News Today calculation indicates a gross margin of roughly 12.1%, compared with about 11.1% in the prior-year period. In a manufacturing business operating at that level of gross profitability, even relatively modest improvements in material selection, yield or production efficiency can become economically meaningful if they are sustained across sufficient production volume.

Could graphite electrode demand make manufacturing optimization increasingly valuable?

Graphite electrodes are essential consumables in electric arc furnace steelmaking because they conduct the electrical current used to melt scrap and other metallic inputs. Electric arc furnaces accounted for roughly 30.3% of global crude steel production in the latest World Steel Association process data, giving graphite electrode producers exposure to a major segment of global steelmaking.

The longer-term industry backdrop is also shaped by steel decarbonization. Scrap-based electric arc furnace production generally has a substantially lower emissions intensity than conventional blast furnace and basic oxygen furnace steelmaking, although actual emissions depend heavily on electricity generation and raw-material inputs. Greater deployment of electric arc furnace technology could therefore support structural demand for graphite electrodes, even though near-term electrode demand remains cyclical and closely linked to steel production.

That cyclical element should not be ignored. World Steel Association expects global steel demand to increase only 0.3% in 2026 before accelerating to 2.2% in 2027. Graphite electrode markets have also continued to face variations in needle coke costs, steel mill utilization and regional purchasing activity. Manufacturing efficiency becomes particularly useful in that environment because producers cannot rely exclusively on higher selling prices to improve profitability.

Predictive formulation technology could give Fangda Carbon another lever to manage those conditions. If the company can switch among suitable feedstocks faster when relative pricing changes, procurement decisions can become more responsive to the market without requiring the same volume of trial-and-error testing. The potential advantage is therefore less about predicting steel demand and more about protecting manufacturing economics when the variables Fangda Carbon can control begin to move.

What does the Fangda Carbon deployment prove about XtalPi’s expansion beyond pharmaceuticals?

XtalPi built much of its identity around artificial intelligence, physics-based computation and robotic automation for pharmaceutical research. The company has since been pushing the same underlying technology architecture into materials science, energy, agriculture and other scientific industries. Fangda Carbon is important because heavy industrial manufacturing tests whether those capabilities can generate value far outside molecular drug discovery.

The timing also matters financially. XtalPi reported first-half 2026 revenue of RMB393.6 million, down from RMB517.1 million a year earlier, largely because the comparison period included a US$51 million upfront payment from a drug-licensing transaction. Excluding that licensing effect, the company said underlying revenue increased 73.8%, while revenue from its AI for Science Intelligent Solutions business increased 136.4%.

XtalPi nevertheless remains in an investment phase. The company reported a first-half net loss of RMB224.9 million compared with a RMB75.6 million profit in the corresponding 2025 period, while research and development expenditure increased 66% to RMB367.8 million. Its balance sheet provides substantial capacity to continue investing, with XtalPi reporting total cash of RMB8.67 billion at June 30, 2026.

That makes industrial deployments particularly relevant to the investment story. XtalPi has spent heavily building autonomous laboratories, scientific models, agentic systems and specialized datasets. Investors ultimately need evidence that those assets can support repeatable commercial contracts rather than remaining primarily research infrastructure. A manufacturing model that can be adapted across additional materials, processes and customers would therefore have greater strategic value than a one-off project built exclusively for Fangda Carbon.

Can the Fangda Carbon model become a reusable industrial AI product rather than a custom project?

The next phase of the partnership provides the clearest indication of XtalPi’s ambition. The companies plan to extend their work from raw-material selection into broader formulation design, process optimization and production feedback, creating a more comprehensive system covering additional stages of carbon-material development and manufacturing.

Their earlier strategic agreement was more ambitious still. XtalPi and Fangda Carbon proposed developing AI models for material design and process optimization, combining digital twins, artificial intelligence and robotics, and building an intelligent research infrastructure for carbon materials. The accepted raw-material model therefore represents an early operational component of a considerably larger architecture.

Scalability will depend on how much of the underlying software can be reused. A model trained heavily on Fangda Carbon’s proprietary historical data may produce substantial value for that company while still requiring extensive customization before deployment at another manufacturer. XtalPi will strengthen the economics of its industrial AI business if common data-engineering, optimization and model components can be redeployed across different customers while company-specific information provides the final layer of specialization.

There is also a potentially important data flywheel. Once an AI model becomes part of an active manufacturing workflow, new production outcomes can be fed back into the system, creating additional examples of successful and unsuccessful formulations. If that feedback improves prediction accuracy over time, operational deployment itself can strengthen the underlying product.

The evidence needed now is commercial rather than conceptual. Additional accepted modules, new industrial customers, repeat contracts and disclosed economic improvements would show whether XtalPi is building a replicable materials intelligence business. Without those indicators, Fangda Carbon remains an encouraging validation case rather than proof of a large new revenue engine.

How are XtalPi and Fangda Carbon shares positioned before investors can react to the AI deployment?

The announcement timing means neither stock provides a clean post-release market reaction yet. XtalPi closed at HK$7.95 on October 2, down 2.9% for that session, but the PR Newswire announcement was released after the Hong Kong market had already closed. The move therefore preceded the announcement and should not be attributed to the Fangda Carbon deployment.

XtalPi shares had nevertheless strengthened over the preceding month, rising from HK$7.20 on September 3 to HK$7.95 on October 2, an increase of about 10.4%. The stock remained well below its 52-week high of HK$15.12, with the latest 52-week range running from HK$6.30 to HK$15.12. That positioning suggests investors have recently shown greater interest in the shares, but the valuation remains substantially below the peaks reached during the past year.

Fangda Carbon’s latest available mainland close was CNY5.29 on September 30, compared with CNY5.21 on September 3, representing an increase of roughly 1.5% over that interval. Its 52-week range stood at approximately CNY4.92 to CNY7.29. Because its latest recorded trading session preceded the October 2 international announcement, there is likewise no basis for describing Fangda Carbon’s share price as having reacted to the AI milestone.

The more useful investor signal will come after trading resumes and, more importantly, through subsequent operating disclosure. A short-term share move can show that the announcement attracted attention, but quantified cost savings, broader production deployment and additional industrial AI contracts would provide stronger evidence of fundamental value creation.

What would show that XtalPi’s graphite electrode AI is creating measurable industrial value?

The accepted model has already passed a more meaningful test than a laboratory demonstration because it is operating within Fangda Carbon’s production workflow. The next step is proving that technical acceptance produces measurable economics.

For Fangda Carbon, useful indicators would include lower raw-material costs per tonne, fewer formulation trials, faster qualification of substitute materials, improved yield or more stable electrode performance when input characteristics change. For XtalPi, the evidence would be broader deployment within Fangda Carbon, follow-on revenue, additional materials-industry customers and increasing contribution from AI for Science Intelligent Solutions.

The partnership also illustrates a larger industrial AI opportunity. Manufacturing companies often possess decades of production information that was created for operational purposes rather than machine learning. Turning those fragmented records into structured datasets and then into predictive systems can potentially unlock value without requiring an entirely new physical production process.

XtalPi now has a live example of that thesis in graphite electrodes. The model does not eliminate physical verification, and the companies have not disclosed enough financial information to quantify its return. What it does establish is that XtalPi’s AI platform can move from scientific prediction into a real manufacturing decision loop, where cost, materials and production quality are directly connected.

The next milestone is therefore not another claim about model capability. It is evidence that Fangda Carbon can convert fewer physical trials, faster substitution decisions and better formulation screening into sustained operating savings, while XtalPi proves that the same architecture can be sold repeatedly across advanced materials manufacturing.

What are the key takeaways from XtalPi and Fangda Carbon’s industrial AI deployment?

• XtalPi Holdings Limited and Fangda Carbon New Material Co., Ltd. have moved a jointly developed raw-material selection AI model into Fangda Carbon’s graphite electrode production workflow.
• The model has passed project acceptance after validation using independent datasets and production trials.
• The system predicts performance and ranks raw-material formulations before physical testing, reducing the number of candidate combinations experts need to validate.
• Final graphite electrode formulations remain subject to human expertise and experimental or production verification.
• The companies have not disclosed the percentage cost savings, production-yield improvement or financial return generated by the model.
• Fangda Carbon reported first-half 2026 revenue of RMB1.97 billion, up 16.6%, while attributable net profit increased to RMB170.9 million.
• XtalPi reported RMB393.6 million of first-half revenue and said AI for Science Intelligent Solutions revenue increased 136.4% year on year.
• XtalPi remained loss-making in the first half of 2026 but reported RMB8.67 billion in total cash at June 30.
• The companies intend to expand the partnership into formulation design, process optimization and production feedback.
• The strongest future proof point will be quantified manufacturing savings and evidence that XtalPi can replicate the industrial AI model across additional materials customers.