iSoftStone SZ 301236 Public

iSoftStone (301236) Deep Dive: A ChiNext-Listed IT Services Giant Whose TianShu Platform Builds a Three-Layer Physical-AI Stack with Self-Developed TwinHub Digital Twin, FSim Industrial Simulation, CAE Multiphysics and Phyxis Embodied Control

iSoftStone (301236, listed on Chinext in 2022) builds a full-stack physical-AI system through its TianShu platform, led since August 2026 by a 1+1+4 architecture with a hybrid world model as the brain, combining self-developed iTwins digital twins, FSim industrial simulation, a CAE toolchain and Phyxis embodied control, and introducing the ground-air integrated robot dog Tianqing F1. But attributable net profit swung to a loss of CNY 115M in the first nine months of 2026 and operating cash outflow reached CNY 3.92B, while physical AI remains separately undisclosed.

Executive Summary

iSoftStone (301236) is a large Chinese IT services and computing products company listed on ChiNext in 2022. First half 2026 revenue was 18.607 billion renminbi, up 17.91 percent. AI-related business reached 11.435 billion or 61.46 percent, up 46.61 percent. Computing products and intelligent electronics reached 9.479 billion or 50.94 percent, up 40.32 percent and passing half of total revenue for the first time, while software and digital technology services were 8.970 billion or 48.21 percent and digital energy and intelligent computing services were 149 million or 0.80 percent.

The profit line needs closer reading. Second quarter revenue was 10.490 billion, up 19.61 percent year over year and 29.22 percent sequentially, with attributable net profit of 85 million, up 54.71 percent year over year and 124.36 percent sequentially, while adjusted net profit was negative 28 million. Gross margin was 11.01 percent and net margin 0.81 percent. The quarterly attributable profit therefore depends on non-recurring items, with the core business still marginally loss-making on an adjusted basis, and an 11 percent gross margin indicates a heavy hardware mix.

The longer record is more telling. Attributable net profit was 973 million in 2022, 534 million in 2023, 180 million in 2024 and 206 million in 2025, falling from near one billion to about 200 million over four years. Revenue over the same period rose from 19.104 billion in 2022 to 35.090 billion in 2025, with 2024 jumping from 17.581 billion to 31.316 billion mainly because of the acquisition of the Tongfang computing business, now branded iSoftStone Huafang. Revenue doubling came through acquisition while profit decline reflects real change in core profitability.

In physical AI terms iSoftStone needs careful framing. In 2026 the company adopted a strategy of integrated software and hardware full-stack intelligence anchored on an AI factory positioning, focusing on AI infrastructure, computing intelligence, scenario intelligence and terminal intelligence, and has described a Tianshu physical AI architecture in three layers along with TwinHub digital twins, FSim industrial simulation, CAE multiphysics and Phyxis embodied control. These are genuine investment directions, but their share of 18.6 billion of revenue is undisclosed. From the revenue structure, most of the 61.46 percent AI-related business is AI servers, liquid-cooled workstations, supernode intelligent computing equipment and AIPC hardware, plus token resale and computing services, rather than physical simulation software.

The 2026 third-quarter report, disclosed on the evening of October 27, rewrites the conclusion. Revenue for the first nine months reached CNY 29.367B, up 15.70 percent, but attributable net profit swung to a loss of CNY 115M against CNY 98.89M of profit a year earlier, a 216.21 percent year-on-year decline. Adjusted net profit was negative CNY 249M, down 319.33 percent, basic earnings per share was negative CNY 0.11 and weighted average return on equity negative 0.88 percent. Third-quarter revenue was CNY 10.760B (up 12.06 percent) with attributable net profit of CNY 149.7M (down 37.99 percent but up 75.7 percent sequentially) and adjusted net profit of CNY 138M, up 601.37 percent sequentially, so the quarter returned to profit yet still cannot fill the CNY 265M loss booked in the first half. More consequential is operating cash flow at negative CNY 3.921B against negative CNY 1.610B a year earlier, down 143.56 percent, driven mainly by stocking: inventory rose to CNY 11.162B at end-September from CNY 7.710B at the start of the year. The company attributes the profit decline to inventory write-down provisions (asset impairment losses of CNY 320M for the nine months against CNY 129M a year earlier) combined with interest expense on new bank borrowings (financial expenses of CNY 219M, up 119.63 percent, of which interest expense was CNY 216M).

The "profit recovery in 2026" judgement therefore held on the interim-report basis and has been reversed on the third-quarter basis: the full year is no longer a recovery but a first nine months in loss of CNY 115M. Only two things support the recovery case, namely that the quarterly loss is narrowing with Q3 back in the black, and that the impairment is a one-off provision.

There is also real progress on the physical AI side. At the 2026 World Robot Conference in August, Tianshu launched a "1+1+4" physical AI system with a hybrid world model as the brain (one hybrid world model, one physical AI technology platform, four categories of embodied robots), and introduced the ground-air integrated robot dog Tianqing F1; it later jointly released the CloudRobo ecosystem solution with Huawei Cloud and co-built an embodied intelligence data foundation, while the Tianshu integrated physical AI platform was selected into the DBC Consulting "2026 Top 50 Physical AI Innovation Enterprises" list at rank 20. These are genuine product and ecosystem advances, yet the company still discloses no revenue, customers or order value for physical AI.

One framing point matters. This is a company with large scale, thin margins, a return to loss in the first nine months of 2026, and physical AI as a small slice. Its strengths are channels, delivery capability, a software and hardware portfolio and positioning in domestic IT substitution and AI computing infrastructure. Its weaknesses are profitability at a 0.81 percent net margin, constraints on software products from a hardware DNA, and the absence of separate disclosure for physical AI. It cannot be assessed on a physical AI company framework but only as IT services plus computing hardware plus an AI platform in transition.

Company Overview

From IT Outsourcing to Integrated Hardware and Software

iSoftStone was founded in 2005 and long provided IT outsourcing and digital transformation services to finance, manufacturing, energy, telecommunications and internet customers. After its 2022 ChiNext listing it expanded through acquisition, most importantly the Tongfang computing business that became iSoftStone Huafang, adding design and manufacturing of servers, workstations and terminals.

That step changed both the revenue structure and the nature of the business. Before, it was a software and services company with higher margins and moderate revenue per head. After, it became a hybrid of software services and hardware manufacturing with doubled revenue and diluted margins. The 2024 jump from 17.581 to 31.316 billion is largely explained by this.

Business Structure in the First Half

Three segments exist: computing products and intelligent electronics at 9.479 billion or 50.94 percent, up 40.32 percent; software and digital technology services at 8.970 billion or 48.21 percent; and digital energy and intelligent computing services at 149 million or 0.80 percent.

The overall picture on the third-quarter basis: total assets reached CNY 37.786B at end-September 2026, up 37.84 percent from the prior year end, with net assets attributable to shareholders of CNY 13.095B, up 26.08 percent. The asset-side expansion came mainly from three things: stocking (inventory from CNY 7.710B at the start of the year to CNY 11.162B), proceeds from a non-public share issuance, and receivables growing with revenue. Liabilities expanded in parallel: short-term borrowings of CNY 10.482B and long-term borrowings of CNY 2.577B at end-September, CNY 13.059B in total; financing cash flow net inflow of CNY 9.005B for the nine months, up CNY 7.675B year on year, and investing cash flow net outflow of CNY 2.859B. This says the scale expansion was bought with debt and inventory, not supported by operating cash flow.

By region, North China contributed 7.212 billion or 47.08 percent, South China 5.119 billion or 33.42 percent and East China 2.987 billion or 19.50 percent. North China near half reflects the Beijing headquarters and concentration of state-owned and government customers. A debt ratio of 61.46 percent is normal for a hardware plus project delivery model but above pure software companies.

Four Capabilities of the AI Factory

The strategy is summarized as an AI factory with four capabilities. AI infrastructure includes participation in the national integrated computing network and a self-built token factory supplying compute and tokens to model vendors, state-owned enterprises and research institutions. Computing intelligence includes mass production of AI servers, liquid-cooled workstations and supernode equipment with joint work on high-performance storage and training and inference optimization alongside Tsinghua University's computer science department and the Beijing Academy of Artificial Intelligence. Scenario intelligence uses front-line delivery engineers plus AI platform support across more than ten industries including finance, manufacturing, steel, agriculture, transport and logistics and retail. Terminal intelligence includes AIPC products from Zhitong International, several certified at the national terminal intelligence L3 level, and Mechanical Revolution gaming laptops among the top three domestically.

The internal logic is a closed loop of compute, hardware, scenarios and terminals: sell compute to model vendors, hardware to enterprises, scenario solutions to industries and terminals to consumers. The advantage is broad customer coverage and cross-selling, and the risk is that every layer has stronger specialists, with cloud vendors in compute, Inspur and H3C in hardware, industry ISVs in scenarios and Lenovo and Asus in terminals.

Technology

The Three-Layer Physical AI Architecture

The disclosed Tianshu physical AI architecture has three layers. L1 is the foundation with compute, data platform and simulation engines. L2 is the platform layer with TwinHub digital twins, FSim industrial simulation and CAE multiphysics. L3 is the application layer with Phyxis embodied control and industry solutions.

The layering is architecturally reasonable: the foundation supplies compute and data, the platform supplies modeling and simulation, and applications supply robot control and industry delivery. The goal is a flywheel of twin, simulation, execution and data return, where simulation generates training data, models execute on real equipment producing new data, and that data improves the simulation.

Maturity varies greatly across the layers. L1 compute and data platforms reflect real capability with compute resources and delivery experience. L2 simulation engines are new capabilities that must compete with vendors holding twenty years of solver depth. L3 embodied control is a frontier without mature industry solutions.

From L1/L2/L3 to "1+1+4": the evolution of the system description

The draft of this report described the Tianshu physical AI architecture in three layers, L1 model, L2 spatial intelligence platform and L3 execution terminal, based on public statements from the first half of 2026. Since August 2026 the company has led with a different description: the "1+1+4" physical AI system released at the World Robot Conference.

  • One hybrid world model as the brain. It fuses the semantic understanding of large models with an industrial physical reasoning framework, so natural-language instructions connect directly to objective physical rules, performing environment understanding and task decomposition in a three-dimensional general industrial space, with all reasoning output checked for physical feasibility. Its data comes from two sources: the self-developed digital twin base and millimetre-level 4D engine generating physically compliant operating-condition samples through CAE simulation in bulk, and virtual interaction producing synthetic real-virtual training data through massive robot trial and error in simulation.
  • One technology platform as physical AI infrastructure. The Tianshu physical AI platform standardizes and packages model capability, supporting API and lightweight portal access and offering reasoning subscription services by token, with the NCP intelligent computing base providing compute scheduling and service metering. The platform carries three core capabilities: the digital twin base, full-stack industrial simulation and embodied real-virtual training.
  • Four categories of embodied robots as execution. Humanoid robots, education and home robots, specialist robots and ground-air coordinated robot dogs, six iSoftStone Tianqing products in total, forming a real-virtual self-evolving loop in which the robot receives a strategy, executes it on site and feeds data back into the world model.

This does not negate the original architecture but reorganises the same capabilities for productisation: the three layers describe capability layering while 1+1+4 describes product packaging, one brain, one platform and four robot bodies. Both descriptions coexist in company materials and some coverage still uses the older framing.

Tianqing F1: the ground-air integrated robot dog

Tianqing F1 is the ground-air coordinated robot dog among the four robot categories of the 1+1+4 system. It first appeared at the 2026 World Robot Conference in August and reappeared at the Nanjing robot skills exhibition in October, where it demonstrated the full sequence of the back hatch opening while walking, the drone taking off autonomously for patrol, and precise return to dock.

Its technical form integrates a biomimetic quadruped with an onboard drone into one intelligent work platform: on the ground the quadruped climbs steps, ditches and rubble to reach terrain wheeled equipment cannot enter; in the air the drone ascends for high-altitude inspection and 3D survey, forming a complementary mesh of wide-area aerial coverage and fine-grained ground work. Application directions cover plant security, field exploration and large-site inspection.

In October 2026 the company jointly released the CloudRobo ecosystem solution with Huawei Cloud and co-built an embodied intelligence data foundation. The Tianshu integrated physical AI platform was also selected into the DBC Consulting "2026 Top 50 Physical AI Innovation Enterprises" list at rank 20, and the Longyuan New Energy digital twin simulation platform that the company co-submitted was named among the "2026 Top 10 New Industrialization Excellent Practice Cases". The company states the system has already landed at a large energy site, where unplanned equipment downtime fell 30 percent and process validation and algorithm development cycles compressed from months to weeks. These effect figures come from company statements and carry no third-party benchmark.

TwinHub and FSim Positioning

TwinHub is a digital twin platform mapping physical assets such as production lines, equipment and campuses into interactive 3D models connected to live data. Its technical core is lightweight models, data integration and real-time rendering, overlapping with the capabilities of Frontop and 51WORLD.

One naming difference needs flagging: in public statements from the first half of 2026 the digital twin platform was called TwinHub, while the 1+1+4 system materials from August 2026 and the October IoT Expo materials describe the digital twin base as the self-developed iTwins, paired with a millimetre-level 4D engine and a new "4D data engine" component, whereas coverage such as Securities Daily in June 2026 refers to the digital twin platform as TwinHub. The company has not publicly explained whether the two are a rename, a version iteration or two separate product lines. This report treats them under the functional description "digital twin base" and does not assert a definite relationship between them.

FSim is an industrial simulation platform intended to provide structural, fluid, thermal and electromagnetic analysis. The barrier here is highest, since solver accuracy, robustness and parallel efficiency require long accumulation and cannot be compensated by platform architecture. Choosing in-house CAE means competing directly with Ansys, Siemens and Dassault plus domestic vendors PERA, INTESIM and Tenfong.

The phrase multiphysics deserves attention. Coupled analysis of fluid-structure, thermal-structure and electromagnetic-thermal is the hardest part of CAE, and INTESIM needed more than a decade to provide both strong and weak coupling. A later entrant cannot reach parity in coupling capability quickly.

Crossing From IT Services to Simulation Software

This is the central question for the physical AI business: can an IT services company build competitive simulation software?

Three favorable conditions exist. Engineering capability accumulated in IT services, including large-scale system integration, data processing and heterogeneous compute scheduling, helps with simulation cloud platforms. Customer relationships across finance, manufacturing and energy allow bundling simulation into digital transformation proposals. And compute resources are in house.

Three unfavorable conditions also exist. Solvers are mathematician and engineer work requiring sustained accumulation of numerical methods, material models and validation cases that IT engineering does not transfer into. Software product refinement requires product management culture while IT services runs on project delivery culture, and the two conflict. Sales logic also differs, since IT services sell headcount and projects on cost-plus pricing while simulation sells licenses and value.

Industry experience suggests IT services companies rarely succeed at software products, not from lack of capability but from misaligned organization and incentives. Whether iSoftStone is an exception depends on running simulation as an independent product line with its own team, metrics and brand rather than as a delivery module.

Technical Shortfalls

First, solver depth. CAE multiphysics is new without public accuracy metrics, validation cases or customer cases.

Second, the physical AI revenue share is undisclosed. Most of the 61.46 percent AI-related business is hardware and compute, with physical AI scale unknown.

Third, maturity of Phyxis embodied control. Embodied intelligence is a frontier without scaled deployments, and progress is unknown.

Fourth, path dependence from the hardware DNA. Organization, incentives and customer expectations are built around delivery, and moving to products requires organizational change.

Product and Commercial Model

Commercial Roles

Computing products and intelligent electronics are the revenue engine at 9.479 billion or 50.94 percent with low margins. Customers are government, state-owned enterprises, education and finance buying under domestic IT substitution and computing infrastructure programs, with intense price competition against Inspur, H3C, Lenovo and Sugon.

Software and digital technology services are the legacy core at 8.970 billion or 48.21 percent with higher margins but slower growth, delivered as projects and staff augmentation to large enterprises in finance, manufacturing and energy.

Digital energy and intelligent computing services are new at 149 million or 0.80 percent, small but directionally clear in computing operations and token services.

Margin Structure and Earnings Quality

An 11.01 percent gross margin is low for IT, where pure software exceeds 60 percent, IT services run 25 to 35 percent and hardware runs 8 to 15 percent. Eleven percent indicates hardware dominates revenue.

A 0.81 percent net margin means 0.81 renminbi of profit per 100 renminbi of revenue. In that thin structure, hardware price wars or a slowdown in compute investment erode profit quickly, which is exactly what happened as attributable net profit fell from 973 million in 2022 to 206 million in 2025 while revenue doubled.

Customer and Industry Coverage

Customers span more than ten industries including finance, manufacturing, steel, agriculture, transport and logistics and retail. The value of that coverage is cross-selling, since a steel customer may need compute, servers, digital twins, simulation and embodied solutions together.

Cross-selling presupposes each product is good enough. If simulation accuracy trails specialists, customers split purchases, buying simulation from Ansys and servers from Inspur, turning full stack into a patchwork.

Joint Innovation and Partnerships

The company built a joint AI systems research center with Tsinghua's computer science department, launched a domestic model and chip open source zone with the Fifth Electronics Research Institute of the Ministry of Industry and Information Technology, and invested strategically in SiliconFlow and Tanweixianlian. The strategic value exceeds financial value, with Tsinghua and the ministry institute providing technical endorsement and policy access while the two investments fill key links in inference infrastructure and DPU networking.

Competitive Landscape

IT Services Rivals

IT services competitors include Chinasoft International, Beyondsoft, hiSoft now part of China Electronics Jinqin and industry integrators. The market is competitive with thin margins, won on scale and relationships. iSoftStone's relative advantages are listed-company funding and brand plus the software and hardware combination, with weaknesses in revenue per head and sensitivity to labor cost.

Computing Hardware Rivals

Server and terminal competitors include Inspur, H3C, Lenovo, Sugon and Huawei's Ascend ecosystem. Competition turns on supply chain access to chips, manufacturing scale and domestic IT credentials. iSoftStone Huafang's relative advantage is the Tongfang brand inheritance and education and government channels, with weaknesses in scale against leaders and dependence on upstream chip supply.

Physical AI Rivals

In physical AI and simulation it faces two groups. Specialized simulation vendors such as PERA, INTESIM, Tenfong, Yundao and ZWSOFT hold two decades of solver accumulation. Digital twin vendors such as 51WORLD, Frontop and Uinnova hold visualization and scenario depth.

iSoftStone's differentiation is full stack plus compute plus relationships, none of which constitutes an advantage in simulation specifically, where customers buy accuracy and validation rather than compute and relationships.

Relationship With Cloud Vendors

Cloud vendors are both partners for compute procurement and competitors selling AI solutions and industry models. The token factory competes directly with cloud model services, while iSoftStone positions as a neutral third party serving multiple model vendors, which is its differentiation.

Financing and Capital Structure

Listing and Capital Activity

Listed on ChiNext in 2022, the company expanded hardware capability through the Tongfang computing acquisition. A 61.46 percent debt ratio is above pure software companies and below heavy manufacturing, consistent with the hybrid model.

Historical Profit Trajectory

Attributable net profit was 973 million in 2022, 534 million in 2023, 180 million in 2024 and 206 million in 2025, down 79 percent over four years while revenue rose 84 percent from 19.104 to 35.090 billion.

The data shows scale-driven rather than profit-driven growth. Acquisition-driven revenue did not convert into profit and low-margin hardware diluted overall profitability, which is the cost of the integrated strategy.

First nine months of 2026: from "profit recovery" back to loss

The "profit recovery in 2026" judgement written into the draft of this report on the basis of the interim report has been reversed on the third-quarter basis. This is the directional correction that matters most in this update.

  • Revenue: CNY 29.367B, up 15.70 percent from CNY 25.384B
  • Net profit attributable to shareholders: negative CNY 115M against positive CNY 99M a year earlier, down 216.21 percent
  • Non-GAAP net loss: CNY 249M against a CNY 59M loss a year earlier, down 319.33 percent
  • Operating cash flow: negative CNY 3.921B against negative CNY 1.610B, down 143.56 percent
  • Gross margin: 10.47 percent against 11.39 percent, down 0.92 percentage points
  • Net margin: negative 0.64 percent against positive 0.14 percent, down 0.78 percentage points
  • Basic EPS: negative CNY 0.11 against positive CNY 0.10
  • Weighted average ROE: negative 0.88 percent, down 1.82 percentage points
  • Asset impairment losses: CNY 320M against CNY 129M
  • Financial expenses: CNY 219M against CNY 100M, up 119.63 percent

Two causes explain the swing, and neither is core operating deterioration. First, inventory write-down provisions pushed asset impairment losses to CNY 320M, driven by stocking to absorb raw material price volatility as AI full-stack demand expanded. Second, new bank borrowings to fund that stocking pushed interest expense to CNY 216M.

But "non-operating factors" does not mean "no pressure". Stocking lifted operating cash outflow to CNY 3.921B and total borrowings to CNY 13.059B, inventory of CNY 11.162B is exposed to hardware price swings or a demand slowdown, and interest expense is fixed. The report also shows R&D spending of CNY 659M for the nine months, down 14.50 percent, which sits in tension with the investment intensity implied by the AI full-stack narrative.

The one clearly positive signal is at the quarterly level: third-quarter revenue of CNY 10.760B was up 12.06 percent, attributable net profit CNY 149.7M was down 37.99 percent year on year but up 75.7 percent sequentially, adjusted net profit CNY 138M was up 601.37 percent sequentially, and gross margin of 11.96 percent was down 0.70 points year on year but up 0.95 points sequentially and about 2.4 points above the first half. The quarter returned to profit and improved sequentially, but still cannot fill the CNY 265M loss booked in the first half.

Conclusion: the 2026 financial narrative is not a profit recovery but a first nine quarters in loss on stocking and impairment, with Q3 stabilizing, and the full year depending on whether impairment pressure clears in Q4.

October 30, 2026: acquiring Guangzhou IDC assets for about CNY 180M

An announcement on October 30, 2026 disclosed that subsidiary Beijing iSoftStone Digital Energy Technology plans to acquire 60 percent of Yuantong Cloud Data Technology (Beijing) for CNY 180M in cash and 60 percent of Beijing Yuntai Shulian for CNY 1. The core assets are the Guangzhou Guanda Road and Guangzhou Ruihe Road data centers, to be consolidated after completion, together with 23,100 tonnes of standard-coal energy assessment quota.

The strategic logic is clear: completing the Greater Bay Area picture of headquarters, scheduling platform and physical base for the AI factory strategy, serving the word "token factory" business. About 70 percent of the target's existing tenants are large internet firms and financial institutions, closely matching agentic inference demand and convertible into compute leasing and token service customers.

The risk is equally clear: the transaction carries no performance commitment and no performance bet, and it constitutes a business combination under common control and therefore forms goodwill subject to annual impairment testing. The announcement itself lists operating risk for the target and goodwill impairment risk. Against a backdrop of CNY 3.921B of nine-month operating cash outflow and CNY 13.059B of total borrowings, this acquisition raises leverage further.

DISCREPANCIES

1. The definition of AI-related business at 61.46 percent. The company reports 11.435 billion or 61.46 percent of revenue as AI-related, but that includes computing products and intelligent electronics at 9.479 billion, token resale, computing services and some software services, with hardware dominating. Some interpretations equate it with AI software or platform revenue. This report states that the definition is hardware and compute heavy while the physical AI share is undisclosed. Source tier: OFFICIAL (interim report definition) differing from interpretation.

2. Attributable profit turning positive while adjusted profit is negative. Second quarter attributable net profit was 85 million, up 54.71 percent, while adjusted net profit was negative 28 million. Book profit comes from non-recurring items. Some coverage headlines significant quarterly profit recovery without noting the adjusted loss. This report presents both. Source tier: OFFICIAL.

3. Revenue doubling against profit halving. Revenue rose from 19.104 billion in 2022 to 35.090 billion in 2025, up 84 percent, while attributable net profit fell from 973 to 206 million, down 79 percent. The divergence mainly reflects the structural shift from acquiring low-margin hardware, yet some interpretations treat revenue growth as evidence of quality growth. This report presents both and explains the structural cause. Source tier: OFFICIAL (annual reports).

4. Acquisition effect in the 2024 revenue jump. Revenue rose from 17.581 to 31.316 billion in 2024, up 78 percent, mainly from the Tongfang computing acquisition. Some interpretations treat it as organic growth. This report flags the acquisition and notes that organic growth cannot be isolated from public data. Source tier: OFFICIAL (filings and announcements) plus inference.

5. Actual scale of the physical AI architecture. The company describes the Tianshu physical AI architecture in three layers along with TwinHub, FSim, CAE multiphysics and Phyxis embodied control. These are product directions without separately disclosed revenue, customers or cases, while some interpretations treat them as established businesses. This report labels them directions and investment rather than realized revenue. Source tier: OFFICIAL (strategy and product statements) against financial data.

6. Industry comparison for the 11.01 percent gross margin. An 11.01 percent gross margin is low for IT, where pure software exceeds 60 percent, IT services run 25 to 35 percent and hardware runs 8 to 15 percent. Some analysis applies an IT services framework and concludes the margin is anomalously low, when the company is already hardware dominant. This report explains the margin through the hybrid structure. Source tier: OFFICIAL (data) against industry benchmarks.

7. Relationship between six AI products and four capabilities. The company refers both to four core capabilities of the AI factory and six AI products and services, which classify along different dimensions of capability versus product and are often conflated in coverage. This report separates the capability layer from the product layer. Source tier: OFFICIAL (company statements).

8. Treatment of associates and strategic investments. The company invested in SiliconFlow and Tanweixianlian and built platforms with Tsinghua and the ministry institute. The value in technology, channels and policy is hard to quantify and does not appear separately in filings. This report labels them strategic cooperation without counting revenue contribution. Source tier: OFFICIAL (announcements) with financial contribution absent.

9. Conflicting conclusions between the interim and third-quarter reporting periods. On the interim basis, second-quarter attributable net profit was CNY 85M with adjusted profit negative CNY 28M, and first-half attributable net profit was negative CNY 265M; on the third-quarter basis, nine-month attributable net profit was negative CNY 115M with adjusted net loss of CNY 249M. Both sets come from company periodic reports and both are true, but they cover different periods: the interim covers only the first half while the third quarter covers nine months. This report uses the third-quarter figures whenever describing current operating conditions and keeps the interim figures as a half-year factual record. Source tier: OFFICIAL (interim report) and OFFICIAL (third-quarter report) in parallel, with the third quarter as the current basis.

10. The "2026 profit recovery" conclusion requires reversal. Some interpretations, reading Q2 attributable net profit turning positive at CNY 85M (up 54.71 percent) and Q3 up 75.7 percent sequentially, describe 2026 as a recovery year. On the third-quarter basis, nine-month attributable net profit was negative CNY 115M, adjusted net profit negative CNY 249M and operating cash flow negative CNY 3.921B, so the full year is not a recovery but a year weighed down by stocking and impairment. Ruling: the third quarter governs, and the characterisation of 2026 changes from "profit recovery" to "loss in the first nine months with Q3 stabilizing". The original "Profit Recovery in 2026" section of this report has been rewritten accordingly rather than layering new data on top of an old conclusion. Source tier: OFFICIAL (third-quarter report).

11. Two parallel descriptions of the physical AI system. Public material from the first half of 2026 used the three-layer architecture of L1 model, L2 spatial intelligence platform and L3 execution terminal; from the 2026 World Robot Conference in August the company has led with the "1+1+4" system of one hybrid world model as brain, one physical AI technology platform and four categories of embodied robots, and October IoT Expo materials describe the digital twin base as the self-developed iTwins with a new 4D data engine component, while coverage such as Securities Daily in June 2026 names the digital twin platform TwinHub. This report explains the two as the same capability set reorganised for productisation, but the company has not publicly stated whether TwinHub and iTwins are a rename or a version iteration, and this report does not treat them as the same product. The 2026 Top 50 Physical AI Innovation Enterprises list and the 30 percent reduction in unplanned downtime at an energy site are third-party ranking or company-stated figures without independent benchmarking. Source tier: OFFICIAL (company statements and expo materials) against third-party ranking conventions.

Risks and Diligence Gaps

First, structurally thin profitability that has swung back into loss in the first nine months of 2026. Nine-month net margin was negative 0.64 percent with gross margin of 10.47 percent and attributable net profit of negative CNY 115M. A 0.81 percent net margin already made profit highly sensitive to hardware pricing and the compute investment cycle, and the third-quarter report shows stocking pushing inventory to CNY 11.162B, impairment losses to CNY 320M and interest expense to CNY 216M. A slowdown in AI compute spending or a fall in raw material prices would pressure inventory write-downs and the interest burden at the same time.

Second, acquisition integration and new goodwill risk. The Tongfang computing deal brought revenue while diluting margin, and synergy in revenue and cost is undisclosed. The Guangzhou IDC acquisition announced on October 30, 2026 likewise carries no performance commitment or performance bet and constitutes a business combination under common control, forming goodwill subject to annual impairment testing, with the company itself listing operating risk for the target and goodwill impairment risk in the announcement.

Third, validation of the physical AI business. TwinHub, FSim, CAE and Phyxis have no separate revenue disclosure, so whether they are real businesses or strategic narrative cannot be judged, and CAE multiphysics faces specialist competition with little near-term advantage.

Fourth, hardware supply chain. Core components in servers and terminals including GPUs, CPUs and storage are subject to upstream supply and export controls, and stability directly determines delivery.

Fifth, customer concentration and collection. Government and state-owned customers dominate with long collection cycles, and a hardware share above half means large receivables.

Sixth, organizational transformation. Moving from project delivery to product operations requires organizational and incentive change, and IT services companies have a low success rate at software products as an industry pattern.

Seventh, valuation framework confusion. The market swings between IT services, computing hardware and AI platform frameworks with very different multiples, which drives share price volatility.

Composite Assessment

Technology scores 6.0. AI infrastructure including compute and the token factory and computing intelligence including AI servers, liquid cooling and supernodes reflect real capability, TwinHub and FSim are reasonably architected, and cooperation with Tsinghua and the ministry institute provides endorsement. Deductions for new CAE capability, unknown solver depth, unclear Phyxis maturity and no quantified physical AI metrics.

Business scores 6.0. Revenue of 18.607 billion, 61.46 percent AI-related business, coverage of more than ten industries and cross-selling across software and hardware constitute scale advantages. Deductions for an 11.01 percent gross margin, 0.81 percent net margin, profit dependent on non-recurring items and no separate physical AI disclosure.

Finance scores 5.0. Listed-company financing access, a manageable 61.46 percent debt ratio and the 0.95 percentage point sequential gross margin recovery in Q3 are strengths. Deductions for nine-month attributable net profit swinging to a CNY 115M loss, adjusted net loss of CNY 249M, operating cash outflow of CNY 3.921B (down 143.56 percent), the CNY 11.162B inventory write-down exposure, the rigid burden of CNY 13.059B of total borrowings and CNY 216M of interest expense, and the no-commitment goodwill risk from the new IDC acquisition; the 79 percent profit decline over four years is unchanged.

Product scores 6.5. Six AI products and services cover compute, hardware, scenarios and terminals, AIPC products are certified at national L3 level and Mechanical Revolution ranks top three in domestic gaming laptops, and TwinHub and FSim add software. Deductions for breadth without depth, unverified simulation competitiveness and commoditized hardware.

Market scores 7.0. AI computing infrastructure and domestic IT substitution are explicit policy directions, enterprise AI adoption demand is real and the company has brand and customers in IT services. Deductions for stronger specialists in every segment, intense price competition and an early physical AI market.

Overall 6.1. This is a large company in transition that returned to loss in the first nine months of 2026 and where physical AI is one small slice: acquisitions took revenue to the 35 billion level, the AI factory links compute, hardware, scenarios and terminals, and the 1+1+4 physical AI system (hybrid world model plus platform plus four robot bodies, with Tianqing F1 as the representative product) closes the loop between simulation and embodiment. Yet profitability is thin at a negative 0.64 percent nine-month net margin, profit fell 79 percent in four years, operating cash outflow reached CNY 3.921B and physical AI has no separate data. The judgment on physical AI is that compute and delivery capability are real and the direction of combining a hybrid world model with CAE simulation and embodied bodies is sound, with stated deployments in high-barrier settings such as energy sites and rail transit; but crossing from IT services and hardware to a physical AI platform requires clearing two thresholds in solver depth and product culture that acquisition or partnership cannot cross quickly.

References

  1. iSoftStone 2026 interim report, disclosed August 26, 2026: revenue 18.607 billion renminbi up 17.91 percent; computing products and intelligent electronics 9.479 billion or 50.94 percent up 40.32 percent; software and digital technology services 8.970 billion or 48.21 percent; digital energy and intelligent computing services 149 million or 0.80 percent; other 8.76 million or 0.05 percent; debt ratio 61.46 percent
  2. Regional structure from the same report: North China 7.212 billion or 47.08 percent, South China 5.119 billion or 33.42 percent, East China 2.987 billion or 19.50 percent
  3. AI business data from the same report: AI-related revenue 11.435 billion, 61.46 percent of total, up 46.61 percent
  4. Second quarter 2026 data: revenue 10.490 billion up 19.61 percent year over year and 29.22 percent sequentially, attributable net profit 85 million up 54.71 and 124.36 percent, adjusted net profit negative 27.55 million, gross margin 11.01 percent, net margin 0.81 percent, operating cash flow per share negative 0.49
  5. Strategy statement from the interim report: integrated software and hardware full-stack intelligence anchored on an AI factory positioning, focusing on AI infrastructure, computing intelligence, scenario intelligence and terminal intelligence; six AI products and services covering intelligent computing services, computing products, token resale, front-line delivery engineers, intelligent terminals and internal token use
  6. AI infrastructure and computing intelligence progress from the interim report: participation in the national integrated computing network and a self-built token factory achieving a zero to one breakthrough; mass production of AI servers, liquid-cooled workstations and supernode intelligent computing equipment; joint work with Tsinghua's computer science department and the Beijing Academy of Artificial Intelligence on high-performance storage and large model training and inference optimization
  7. Scenario and terminal intelligence from the interim report: front-line delivery engineers plus AI platform support across more than ten industries including finance, manufacturing, steel, agriculture, transport and logistics and retail; AIPC products certified at national terminal intelligence L3 level; Mechanical Revolution among the top three in domestic gaming laptops
  8. Technology and ecosystem cooperation from the interim report: joint AI systems research center with Tsinghua's computer science department; domestic model and chip open source zone with the Fifth Electronics Research Institute of the Ministry of Industry and Information Technology; strategic investments in SiliconFlow and Tanweixianlian; establishment of the iSoftStone Digital Intelligence Group with a capability system of front-line delivery plus three platforms plus model and compute supply
  9. Honors and market position from company announcements: Fortune China Tech 50 in 2026 and 412th in the Fortune China 500; top domestic vendor in CCID's China IT services and IT consulting rankings; IDC China AI 50 in 2026; top five in IDC China AI professional services with platform and application solutions fourth and infrastructure integration fifth; iSoftStone Computer third in government and education notebook shipments and fourth in desktops in the first quarter of 2026
  10. Historical results from third-party financial data: 2022 revenue 19.104 billion with attributable net profit of 973 million; 2023 revenue 17.581 billion with 534 million; 2024 revenue 31.316 billion with 180 million; 2025 revenue 35.090 billion with 206 million
  11. Industry background from public sources: IT services gross margins typically run 25 to 35 percent, pure software above 60 percent and hardware manufacturing 8 to 15 percent; server and terminal leaders include Inspur, H3C, Lenovo, Sugon and Huawei's Ascend ecosystem; cloud vendors both supply compute and sell industry solutions, creating coopetition with neutral third parties
  12. Technical background from public sources: CAE multiphysics coupling divides into weak partitioned iteration and strong monolithic solving, with strong coupling harder to implement; solver accuracy, robustness and parallel efficiency depend on long accumulation and IT services engineering capability cannot transfer directly; the core of a digital twin platform is model lightweighting, data integration and real-time rendering, a different technical stack from simulation solvers
  13. iSoftStone 2026 third-quarter report, disclosed on the evening of October 27, 2026: revenue for the first nine months of CNY 29.367B up 15.70 percent, net loss attributable to shareholders CNY 115M against CNY 98.89M of profit a year earlier down 216.21 percent, non-GAAP net loss CNY 249M down 319.33 percent, operating cash flow negative CNY 3.921B against negative CNY 1.610B down 143.56 percent, basic EPS negative CNY 0.11, weighted average ROE negative 0.88 percent, gross margin 10.47 percent down 0.92 percentage points, net margin negative 0.64 percent; Q3 revenue CNY 10.760B up 12.06 percent, attributable net profit CNY 149.7M down 37.99 percent but up 75.7 percent sequentially, adjusted net profit CNY 138M down 33.29 percent and up 601.37 percent sequentially, single-quarter gross margin 11.96 percent down 0.70 points year on year and up 0.95 points sequentially; nine-month asset impairment losses CNY 320M against CNY 129M, financial expenses CNY 219M up 119.63 percent including interest expense of CNY 216M, R&D expenses CNY 659M down 14.50 percent, period expenses CNY 2.95B with an expense ratio of 10.05 percent down 0.93 points; total assets CNY 37.786B at end-September up 37.84 percent from the prior year end, net assets attributable to shareholders CNY 13.095B up 26.08 percent, inventory CNY 11.162B against CNY 7.710B at the start of the year, short-term borrowings CNY 10.482B and long-term borrowings CNY 2.577B for CNY 13.059B of total borrowings; financing cash flow net inflow CNY 9.005B and investing cash flow net outflow CNY 2.859B; first-half attributable net loss CNY 265M
  14. Tianshu "1+1+4" physical AI system and Tianqing F1, 2026 World Robot Conference, August 19 to 23, 2026: one hybrid world model as brain, one physical AI technology platform, four categories of embodied robots across six Tianqing products; the ground-air coordinated robot dog Tianqing F1 integrating a biomimetic quadruped with an onboard drone as a debut product; self-developed iTwins digital twin base with a millimetre-level 4D engine, a CAE toolchain plus FSim full-stack simulation platform, and integration of Isaac Sim robot simulation capability; company-stated figures that unplanned equipment downtime at a large energy site fell 30 percent and process validation and algorithm development cycles compressed from months to weeks
  15. Third-quarter-report accompanying disclosure and ecosystem progress, October 28, 2026 and around the same period: joint release of the CloudRobo ecosystem solution with Huawei Cloud and co-building of an embodied intelligence data foundation; the Tianshu integrated physical AI platform selected into the DBC Consulting "2026 Top 50 Physical AI Innovation Enterprises" at rank 20; the co-submitted Longyuan New Energy digital twin simulation platform named among the 2026 Top 10 New Industrialization Excellent Practice Cases; joint construction with industry and academic partners of the AI plus rail embodied intelligence common technology Hebei provincial pilot base; establishment of a physical AI product certification and evaluation laboratory
  16. iSoftStone announcement on a subsidiary's proposed asset purchase, October 30, 2026: wholly owned subsidiary Beijing iSoftStone Digital Energy Technology plans to acquire 60 percent of Yuantong Cloud Data Technology (Beijing) for CNY 180M and 60 percent of Beijing Yuntai Shulian for CNY 1, with the core assets being the Guangzhou Guanda Road and Guangzhou Ruihe Road data centers consolidated after completion and 23,100 tonnes of standard-coal energy assessment quota obtained; the transaction sets no performance commitment and no performance bet, constitutes a business combination under common control, and forms goodwill subject to annual impairment testing
This report is compiled from public sources for informational purposes only and is not investment advice. Financial figures follow company periodic reports; revenue and customers for physical AI products including the Tianshu architecture, TwinHub, FSim, CAE and Phyxis are not disclosed separately, and related judgments are analytical and labeled as uncertain.

Related reports66

This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of the publication date; please refer to the company’s official filings.