Suochen Technology: The Pain of Becoming a Physics-AI Platform from a CAE Vendor

Executive Summary

Suochen Technology is one of the A-share listed industrial software companies with the widest in-house multi-physics solver coverage, and one of the earliest to write "physical AI" into its corporate vision and fundraising plan. Its scarcity does not lie in commercial scale but in how much of the underlying stack it owns: structural, fluid, electromagnetic, acoustic, optical, materials and multi-physics solvers are all developed in house, with the full technical chain held in its own hands.

That scarcity is being diluted by two things.

First, a mismatch in the quality of growth. In H1 2026 revenue reached CNY 72.74M, up 26.83% year over year, while the net loss attributable to shareholders widened to CNY 78.66M, an increase of CNY 32.96M. Over the same period R&D spending reached CNY 84.81M, equal to 116.59% of revenue. Revenue is growing, but cost is growing faster behind every unit of it.

Second, the window for physical AI is narrowing. Ansys, the global CAE leader, completed its acquisition by Synopsys on July 17, 2025, and Altair has also been absorbed into Siemens. Two independent champions disappeared within the same year, which means the consolidation and M&A window for standalone physical-AI platforms is closing.

Our view: high technology score, low commercialization and funding scores. This is a company whose products already work but whose revenue model does not yet. It is worth studying as a technology-route case and as a case study in the sustainability of its burn rate.

Company Overview

Shanghai Suochen Information Technology Co., Ltd. (SH:688507) is listed on the SSE STAR Market. Chen Hao is the actual controller, chairman and general manager, holding 27.58% directly and indirectly as of Q1 2025. The company describes its mission as "exploring physical AI to build a parallel world of physical and virtual reality".

The management background is telling: several core executives previously worked at ANSYS and Asia-Pacific CAE vendors. That history directly explains the product choices that followed — Suochen took the heavy route of getting the solvers right first, before layering AI on top.

The company now runs two product lines:

• Tiangong — CAE industrial software covering structural, fluid, electromagnetic, acoustic, optical, materials and multi-physics analysis.
• Kaiwu — the physical-AI technology base and solutions, including the Intelligent Workshop (surrogate model training), the World Physics Model, physics-based synthetic data, and a physical-AI low-altitude safety platform.

The business has extended from product design and simulation verification into industrial control and production operations management, largely through two acquisitions completed in 2025: Beijing ForceControl YuTong Technology and Kunyu Lancheng (Beijing).

Technology Analysis

Where the moat actually sits

Suochen's moat is not in "AI" — it is in the solver kernel. This is the point most likely to be misread when evaluating the company.

The entire physical-AI narrative (synthetic data, surrogate models, world models) eventually lands on a physics engine. If the engine is inaccurate, synthetic data is biased data, and a surrogate model amplifies the bias. A world physics model is even more direct: it claims to output trustworthy inferences, so the accuracy of the grid, discretization scheme, boundary condition handling and wall modeling is a floor that cannot be avoided.

Suochen has two verifiable assets at that position.

First, breadth of disciplinary coverage. The company develops and iterates its own solvers for fluid, structural, electromagnetic, optical and acoustic analysis. All five shipped new versions in 2025. The general fluid solver introduced a brand-new LBM solver, and the general electromagnetic analysis software WIPL-D covers aviation, aerospace, shipbuilding, power, electronics, semiconductors, computing and communications. The release cadence is one minor version per discipline per quarter and one major version per year. The LBM choice deserves particular attention: the lattice Boltzmann method maps naturally onto parallel GPU execution, which makes it a technical prerequisite for physical-AI data generation rather than an isolated CFD feature.

Second, a four-layer in-house stack. The company describes it as: Tiangong solvers (physics simulation) to the Intelligent Workshop (physical-AI surrogate model training, compressing hours-long simulation into millisecond-scale inference) to the Kaiwu platform (digital twin construction and reinforcement-learning iteration) to the Dasheng agent (a closed loop of perception, prediction, planning, execution and monitoring). All four layers are built by the company itself, keeping the technical chain fully under its own control.

How mature is the stack

It is important to separate shipped products from roadmap.

Shipped as of H1 2026:

• Solver layer: Tiangong solvers across all disciplines, plus LBM and WIPL-D
• Surrogate layer: Intelligent Workshop
• Data layer: Xingzhe, which generates visual, force and acoustic training data through multi-physics simulation
• Model layer: Yingzao Wanxiang, the World Physics Model, released March 2026
• Agent layer: Dasheng, which drives modeling, meshing, solver execution and result analysis through natural language

Roadmap: a general release of Yingzao Wanxiang in 2026, industry world models for urban low-altitude and embodied AI in 2027, and an open ecosystem for secondary development in 2028.

The Dasheng product form deserves separate mention: letting engineers drive modeling, meshing, solving and result analysis through natural language lowers the operational barrier of CAE from "professional software" to "conversation". If the usability holds, its commercial value may exceed that of the underlying solvers, because it directly widens the reachable user base. That said, no public quantitative data exists on this line — no disclosed customer count, no task completion rate — so the 8.0 technology score here does not treat it as scale-validated.

Experiment and validation infrastructure

The 40,000 m² physical-AI facility in Jiaxing houses a numerical wind tunnel, an anechoic chamber, an HPC laboratory and a low-altitude economy laboratory, providing integrated simulation design, virtual verification, physical testing and intelligent optimization. Such physical test facilities are rare among comparable domestic companies. They give the simulation-to-measurement loop a physical anchor, and they are one of the preconditions for physics-based synthetic data to be called high fidelity.

On August 26, 2026, controlling subsidiary Jiaxing Suochen signed a strategic cooperation agreement with AVIC Test & Validation (Beijing) to build a joint laboratory for commercial aerospace and high-end equipment, providing one-stop testing, simulation verification, AI-assisted design and launch safety evaluation across the Yangtze River Delta. This is a typical move of extending simulation capability toward both verification and testing.

Products and Deployment

Product matrix and revenue structure

The company has disclosed a proposed investment project, "Physical Synthetic Data Core Technology Base Construction and Multi-domain Scenario Demonstration Application", with a three-year construction cycle covering compute equipment, hardware-software integration, core R&D, experiment platform build-out and scenario demonstration. That the fundraising project is named directly after synthetic data confirms how high physical AI sits in the company's priorities.

By product line in H1 2026:

• Tiangong industrial automation software: CNY 27.50M, 37.8% of revenue, 85.55% gross margin
• Tiangong engineering simulation software: CNY 14.81M, 20.36%, 58.15% gross margin
• Tiangong integrated production control solution: CNY 10.06M, 13.83%, 14.53% gross margin
• Kaiwu simulation product development: CNY 6.59M, 9.06%, 76.72% gross margin
• Tiangong simulation product development: CNY 6.19M, 8.52%

One structural fact needs to be stated plainly: the physical-AI line, Kaiwu, contributed only CNY 6.59M in H1, or 9.06% of revenue. Industrial automation software from the acquisition accounted for 37.8% on its own. In other words, the company's current main revenue source is not physical AI — it is industrial automation software obtained through M&A.

The comparison is even clearer for full-year 2025: Tiangong CAE revenue reached CNY 284M, 61% of total, of which engineering simulation software was CNY 198M and simulation product development CNY 87M; the production control and industrial automation software contributed by ForceControl together came to CNY 79.31M, or 17%.

Physical AI therefore means, for Suochen today, something closer to a narrative upgrade and a re-rating than to revenue substitution. The Kaiwu gross margin of 76.72% does confirm it is a high-value business, but a base of CNY 6.59M cannot yet carry an independent commercial curve.

Target customers and industries

The company serves aerospace, shipbuilding, ordnance, electronics, automotive and energy with simulation software, simulation product development and technical services. Its four priority directions are specialized industries, commercial aerospace, the low-altitude economy and embodied AI. All four share the same engine — the world physics model.

The low-altitude direction is the most notable case: in June 2025 the company signed a strategic framework agreement with the Transport Bureau of Yuecheng District in Shaoxing to jointly advance a low-altitude physical-AI platform. Rather than merely selling software, Suochen builds a multi-dimensional physical asset — a dynamically generated real-time low-altitude multi-physics map supplying high-precision data for UAV path planning, flight control and airspace risk monitoring.

This "selling physical assets rather than software licenses" model would, if it could be replicated in low-altitude and embodied AI, deliver materially better business quality than traditional CAE licensing. For now, Yuecheng is the only publicly verifiable case, and a single sample is not enough to judge replicability.

Business Model

Revenue structure

Suochen's revenue has four distinct components with different charging logic:

1. Software licenses and subscriptions under Tiangong engineering simulation — the traditional high-margin model, 86.33% gross margin in 2025.
2. Simulation product development under Tiangong and Kaiwu — delivered per project, at 38.79% gross margin for Tiangong and 37.20% for Kaiwu in 2025, far below software.
3. Integrated production control solutions from ForceControl — project-based, with only 14.53% gross margin in H1 2026, typical of systems-integration economics.
4. Industrial automation software from ForceControl — 85.55% to 91.60% gross margin, strongly software-like.

Key observation: weighting the four gives a blended gross margin of 60.25% in H1 2026, against 68.16% for full-year 2025. The 7.91 percentage point decline is not a technical regression. It reflects the rising share of low-margin integration revenue — in other words, the revenue growth brought by acquisitions is of lower quality than organically grown software revenue.

Seasonality

Revenue recognition is heavily concentrated in Q4, and only CNY 72.74M was recognized in H1 2026. This has two consequences. First, interim results look far worse than the full year by construction. Second, cash flow is badly mismatched with revenue, and a negative operating cash flow in the first half is structural rather than a sign of operational deterioration.

The metric more worth watching than the income statement is that H1 2026 operating cash flow was negative CNY 38.35M, an improvement of CNY 25.78M over the negative CNY 64.13M a year earlier. Collections are improving.

Receivables: the number that must be watched

As of end-June 2026, the carrying balance of accounts receivable was CNY 951M, down CNY 93M from the start of the period.

CNY 951M of receivables stands against FY2025 revenue of CNY 466M — receivables exceed twice the annual revenue. Meanwhile, net assets attributable to shareholders were CNY 2.71B and total assets CNY 3.33B, so receivables account for close to 30% of total assets.

This structure is not unusual for a business whose customers are predominantly specialized industries, ordnance and aerospace, where payment cycles are long and acceptance milestones are many. It also means, though, that the company's growth is substantially funded on its own balance sheet. Whether the improvement in collections (CNY 93M reduction) continues matters more for cash flow quality than any single quarter's revenue growth rate.

Competitive Landscape

Globally: two independent champions vanish in one year

2025 was a turning point for the CAE industry:

• On July 17, 2025, Synopsys completed its acquisition of Ansys, giving the combined company an expanded addressable market of roughly USD 31 billion. Ansys common stock ceased trading on Nasdaq, and its former CEO Ajei Gopal and former director Ravi Vijayaraghavan joined the Synopsys board. Ansys 2026 R1 has since shipped under the new owner.
• Altair has also been acquired by Siemens, with HyperWorks folded into Simcenter, HPCWorks into Siemens HPCWorks, and RapidMiner becoming part of Siemens RapidMiner.

The upshot is that the top two independent CAE vendors disappeared in the same year. Consolidation in physical AI is no longer a possibility but a fait accompli. Suochen's window is correspondingly narrower: the time left for standalone physical-AI platforms is shorter than it was two years ago.

Domestically: position and weaknesses

Among Chinese CAE vendors, Suochen sits in the group with leading in-house depth and clearly trailing scale. Pressure comes from two directions.

Downward, at the tool layer — the market also contains solutions that build engineering and industry adaptation on top of open-source solvers such as OpenFOAM and FEniCS, delivering faster and cheaper, creating price pressure in some general-purpose scenarios.

Upward, at the platform layer — the CAE majors are making physical AI their main battlefield. The post-Ansys Synopsys roadmap explicitly targets multi-physics fusion across the full EDA stack, with the first integrated capabilities planned for H1 2026. Suochen must face the fact that Ansys has a natural entry point through its verification and testing business, Siemens has the Xcelerator platform and an industrial customer base, and for both of them physical AI is an incremental feature for existing customers. At Suochen it is a standalone product line built from zero.

That is why this report assigns a market position score of 5.5: a solid technical starting point, but distribution channels and customer base are not in the same league.

Where Suochen differentiates

Suochen's differentiation is not a feature checklist but a delivery model. The combination of a four-layer in-house stack, physical test facilities and a data service offering (DaaS) allows it to deliver data, model and validation as one integrated package rather than selling software licenses. If that path works, the business quality is meaningfully higher than traditional CAE.

But the path demands a different kind of organizational capability: moving from a project-delivery software company to a platform product company. H1 2026 sales expenses rose 151.88% and administrative expenses rose 53.08%. The company is squarely in the cost-leadership phase of that transition.

Financials and Funding

FY2025 (audited)

• Revenue: CNY 465.8M, up 22.97% from CNY 378.8M
• Net profit attributable to shareholders: CNY 31.50M, down 24.00% from CNY 41.45M
• Non-GAAP net profit: CNY 27.26M, down 25.47%
• Operating cash flow: CNY 60.21M, turning positive from negative CNY 48.81M
• Gross margin: 68.16%, down 3.70 percentage points
• R&D as a share of revenue: 23.41%, down 5.08 percentage points
• Basic EPS: CNY 0.36, down 23.40%
• Weighted average ROE: 1.10%, down 0.35 percentage points
• Total assets: CNY 3.47B, up 13.04%
• Net assets attributable to shareholders: CNY 2.86B, up 0.05%

2025 was a year of divergence between revenue and profit: revenue grew nearly 23% while profit fell 24%, because the enlarged consolidation scope raised fixed costs before revenue scaled to match.

H1 2026

• Revenue: CNY 72.74M, up 26.83% from CNY 57.35M
• Net loss attributable to shareholders: CNY 78.66M, versus a CNY 45.70M loss a year earlier
• Non-GAAP net loss: CNY 81.22M, versus CNY 48.01M
• Operating cash flow: negative CNY 38.35M, improved from negative CNY 64.13M
• Gross margin: 60.25%, up 18.36 percentage points
• Net margin: negative 131.88%
• R&D spending: CNY 84.81M, up 67.53%, equal to 116.59% of revenue
• Selling expenses: CNY 29.49M, up 151.88%
• Administrative expenses: CNY 61.12M, up 53.08%
• Basic EPS: negative CNY 0.89, versus negative CNY 0.52

Attributing the loss

Two components of the H1 2026 loss need to be separated.

The first is explicit drag from the acquisitions. ForceControl and its subsidiaries generated CNY 38.08M of revenue and a net loss of CNY 25.62M; Kunyu Lancheng generated CNY 4.65M of revenue and a net loss of CNY 5.20M. Together they lost CNY 30.82M.

The second is the front-loaded cost of in-house investment. R&D spending of CNY 84.81M was fully expensed and equal to 116.59% of revenue. This is a deliberate choice — the physical-AI platform, the synthetic data base and the Dasheng agent are all in their investment phase, and the income statement must bear that.

Conclusion: looking only at the CNY 81.22M non-GAAP loss understates operational improvement — operating cash flow improved by CNY 25.78M, gross margin rose 18.36 percentage points (72.23% in Q2 alone), and collections improved with receivables down CNY 93M. Looking only at those signals, however, understates the burn pressure — R&D at 116.59% of revenue and CNY 951M of receivables mean the company's self-generated cash flow cannot yet support the current pace.

Funding capacity

As a STAR Market listed company, Suochen has an equity financing platform and completed its external expansion through acquisitions in 2025. Net assets attributable to shareholders of CNY 2.71B and total assets of CNY 3.33B leave the balance sheet structure sound. But the combination of persistent losses, high receivables and negative operating cash flow means it faces a clear refinancing requirement within the next 12 to 18 months if it maintains the current investment intensity. At that point the financing environment and valuation level will directly determine the pace of its physical-AI strategy.

Risks and Challenges

Technology risk

The core risk for physical AI is whether the credibility of synthetic data can win over the engineering community. Industrial customers, especially in aerospace and ordnance where error tolerance is near zero, treat simulation results as binary — either fully trustworthy, or not used at all. Suochen's answer is the combination of in-house solvers and physical test facilities, which is methodologically sound, but the Yingzao Wanxiang world physics model was only released in March 2026, the general version has not shipped, and industry models are not due until 2027. There is a lag of at least two years between technology validation and commercialization.

A second risk is the compute ecosystem. Sell-side research has explicitly noted that international giants retain clear advantages in GPU compute ecosystem and global channels, and that Suochen still needs to strengthen compute-ecosystem collaboration. Physical AI is compute-intensive, and Suochen's bargaining power over supercomputing and GPU clusters is one of the key variables determining whether it can compete with Ansys.

Commercial risk

The revenue structure is the most immediate problem. As shown above, Kaiwu contributed only 9.06% of H1 revenue while acquired industrial automation software contributed 37.8%. The company tells a physical-AI story publicly while the revenue structure tells a different one. That mismatch is a reasonable short-term choice for monetizing assets, but if physical AI fails to scale, the market will price the company as an industrial software or systems-integration vendor rather than a physical-AI platform.

Seasonality deserves attention too. With revenue concentrated in Q4 recognition, any delay in acceptance of a major project produces a visible impact on that year's numbers.

Financial risk

CNY 951M of receivables is a double-edged instrument: it reflects one side of a high-barrier customer base, and the other side is capital tied up. Non-GAAP losses on top of negative operating cash flow mean dependence on external funding is rising.

Policy risk

Import substitution is Suochen's most certain tailwind — the compliance and export-restriction environment for international giants in China has, if anything, opened room for domestic CAE. At the same time, new markets such as the low-altitude economy and embodied AI are themselves policy-dependent, so shifts in policy pace transmit directly into the pace of new business deployment.

Overall Assessment

• Technology 8.0 — full in-house multi-physics solver coverage, a four-layer in-house physical-AI stack, and physical test facilities. Held back by the lack of scale validation for the world physics model and weaker compute ecosystem than international peers.
• Commercialization 5.0 — revenue keeps growing while profit moves the other way, gross margin diluted by acquisitions, and physical AI contributing only 9.06% of revenue.
• Funding 5.5 — a STAR Market listing platform, CNY 2.71B of net assets and demonstrated M&A capability, offset by persistent losses, CNY 951M of receivables and rising refinancing pressure.
• Product maturity 7.5 — Tiangong is commercially deployed across all disciplines with a clear quarterly iteration cadence, while Kaiwu is still in product-definition stage despite a dense new release slate.
• Market position 5.5 — first tier among domestic in-house CAE vendors, but clearly behind on scale and global channels, under dual pressure from the Ansys/Synopsys and Siemens consolidations.

References

1. Shanghai Suochen Information Technology Co., Ltd. 2025 Annual Report, April 28, 2026, Shanghai Stock Exchange
2. Shanghai Suochen Information Technology Co., Ltd. 2026 Half-Year Report Summary, August 26, 2026, Shanghai Stock Exchange
3. Suochen investor relations activity record 2026-002, July 21, 2026
4. Suochen investor relations activity record 2026-003, August 25, 2026
5. Synopsys Completes Acquisition of Ansys, July 17, 2025
6. Suochen official product and news pages, demxs.com
7. China Securities Journal, "Product Supply and Demand Both Booming, Listed Companies in the Non-ferrous Sector Report Results", April 14, 2026

This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of October 21, 2026; please refer to the company's official filings.