HollyWave SSE 688682 Public

HollyWave (688682) Deep Dive: RDSim AI Electromagnetic Simulation and Engineering Agents Close the Simulation-Measurement-Product Loop

HollyWave (688682) links simulation, measurement and products via RDSim AI 2026 electromagnetic simulation (AI surrogate plus engineering agent); 2026 first nine months revenue RMB 183 million, down 14.93 percent, with attributable net profit of RMB 10.99 million, up 56.06 percent.

Executive Summary

HollyWave, formally Shanghai HollyWave Electronic System Technology Co., Ltd. and listed on the Shanghai Stock Exchange Star Market under 688682, is a company that places electromagnetic simulation, electromagnetic measurement and phased array products inside one engineering system. It differs from pure CAE software vendors: of 303 million renminbi of revenue in 2025, electromagnetic measurement systems contributed 263 million renminbi or 86.67 percent, while the electromagnetic field simulation and verification business contributed only 28.9 million renminbi or 9.53 percent, and phased array products contributed 0.66 million renminbi or 0.22 percent. That structure means the first thing to establish is that this is a measurement systems company by cash flow and an electromagnetic CAE software company by strategic direction.

2025 was a turning point. Revenue reached 303 million renminbi, up 12.72 percent, while net profit attributable to shareholders swung from a 14.43 million renminbi profit in 2024 to a 23.39 million renminbi loss. Net profit excluding non-recurring items was a loss of 29.85 million renminbi against a 2.61 million renminbi profit in 2024. Gross margin fell to 24.77 percent, down 14.20 percentage points, with the measurement systems segment at 16.39 percent, down 18.31 percentage points. The company attributes the deterioration to the market environment affecting measurement systems and to higher research spending on AI plus CAE. Operating cash flow was positive 83.74 million renminbi, up 379.25 percent, a sharp contrast with the reported loss.

The first half of 2026 showed improvement without full recovery. Revenue was 118 million renminbi, up 15.89 percent, and net profit attributable to shareholders was 4.38 million renminbi, up 1,194.50 percent. Net profit excluding non-recurring items remained a loss of 3.25 million renminbi, narrowing by 1.04 million from a 4.28 million loss a year earlier. Operating cash flow was negative 21.20 million renminbi against negative 48.64 million a year earlier. The headline profit growth rests on non-recurring items totaling 7.62 million renminbi, including 6.50 million in government subsidies.

The third quarter report of October 30, 2026 introduced a divergence. Revenue for the first nine months was 183 million renminbi, down 14.93 percent, turning the first half growth of 15.89 percent into a decline. Attributable net profit was 10.99 million renminbi, up 56.06 percent, and net profit excluding non-recurring items turned positive at 1.00 million renminbi, up 171.16 percent from a loss. Third quarter revenue alone was 64.44 million renminbi, down 42.85 percent. Falling revenue and rising profit point to gross margin: margin for the first nine months was 30.92 percent, up 5.11 percentage points, and the third quarter alone reached 39.61 percent against 26.19 percent in the first half. Research spending was 35.29 million renminbi, 19.30 percent of revenue. The figures describe a company under delivery and volume pressure whose earnings quality is nonetheless improving.

From a physical AI standpoint there are three levers, and they differ from those of other companies. The first is the electromagnetic domain itself: antennas, phased arrays, radar cross sections and electromagnetic compatibility are expensive to solve and have large design spaces, which makes them natural candidates for surrogate acceleration. The second is RDSim AI 2026, the three-dimensional electromagnetic simulation release published on May 25, 2026, whose stated capabilities are AI surrogate models, engineering agents and complex-scenario simulation. The third is a simulation-to-measurement loop that few competitors can replicate, because the company also sells the measurement systems that generate real antenna data for calibrating simulation models.

One framing point matters. The physical AI content is not yet material to the revenue structure. Simulation is under one tenth of revenue, AI features are not separately priced, 2025 was a loss year on an operating basis, and although adjusted profit turned slightly positive in the first nine months of 2026, the net margin on adjusted profit is only 0.55 percent. The value lies more in the scarcity of a combined domestic electromagnetic CAE and physical calibration capability than in realized AI revenue.

Company Overview

Positioning and Business Boundaries

HollyWave has long focused on electromagnetic technology and product development, with three segments: electromagnetic field simulation and verification, electromagnetic measurement systems, and phased array products. Customers span electronic information, aviation, aerospace, shipbuilding, commercial space and the low-altitude economy. Mapped to the engineering workflow, the three segments cover design, verification and delivery: simulation software handles analysis and optimization during design, measurement systems verify performance on prototypes and in production, and phased array products deliver antennas and arrays directly.

This combination of software, equipment and products is unusual among Chinese CAE companies. Most domestic CAE vendors sell software only, leaving verification to third-party chambers or customer-built test ranges. HollyWave makes measurement systems its main business, which puts the entry point of the verification step under its own control. Commercially that means steadier revenue from high-unit-price project deliveries; technically it means a more complete data loop.

Revenue Structure in 2025

By product, electromagnetic measurement systems generated 263 million renminbi in 2025, up 11.99 percent, accounting for 86.67 percent of revenue at a 16.39 percent gross margin, down from 34.70 percent in 2024, a decline of 18.31 percentage points. Electromagnetic field simulation and verification generated 28.9 million renminbi or 9.53 percent. Phased array products generated 0.66 million renminbi or 0.22 percent. Other revenue was 10.86 million renminbi or 3.58 percent. All revenue was domestic.

These numbers deserve two readings. The first is growth: measurement systems grew 11.99 percent and drove total revenue growth, indicating better delivery execution. The second is margin: measurement system gross margin halved from 34.70 to 16.39 percent and dragged overall profitability down. The annual report attributes this to changes in the industry market environment, a formulation that does not specify whether the cause was bidding pressure, customer mix, a higher share of self-manufactured hardware or a shift in cost structure.

Research Spending and Staffing

Research spending in 2025 was 57.82 million renminbi, up 37.05 percent, equal to 19.07 percent of revenue, with a five-year compound growth rate of 13.32 percent. Research staff numbered 80, up 3.90 percent from 77, representing 46.24 percent of total employees versus 44.25 percent previously, with average research compensation of 426,200 renminbi including share-based payment.

Supporting three product lines with 80 researchers sets a high bar for output per person. Against 303 million renminbi of revenue and roughly 173 total employees, revenue per employee is about 1.75 million renminbi, mid-to-upper among Chinese industrial software companies, though software-only productivity cannot be isolated because hardware and system integration are included.

Operating Changes in the First Half of 2026

First half revenue was 118 million renminbi, up 15.89 percent, with second quarter revenue of 38.40 million renminbi, up 32.16 percent year over year but down 52.0 percent sequentially. Attributable net profit was 4.38 million renminbi, up 1,194.50 percent, while second quarter attributable net profit was negative 2.11 million renminbi, an improvement of 44.51 percent from a 3.80 million loss a year earlier and a sequential decline of 132.54 percent. Non-recurring-adjusted net profit was a 3.25 million renminbi loss for the half, with the second quarter loss widening 28.77 percent year over year to 8.29 million renminbi.

Two conclusions follow. Revenue is highly seasonal, with the first quarter typically the peak for acceptance-based recognition and the second quarter falling by more than half, a rhythm tied to the annual budget and acceptance cycles of defense and aerospace customers. And the operating basis has not turned positive, since the second quarter adjusted loss widened to 8.29 million renminbi, meaning the half-year book profit depends on government subsidies and gains on financial assets.

At the end of June 2026 total assets were 864 million renminbi, down 8.9 percent from the prior year end, and attributable net assets were 595 million renminbi, down 2.6 percent. The balance sheet is conservative, but the decline in total assets needs to be read with cash flow: in 2025 cash fell 56.86 percent while trading financial assets rose 27.41 percent, indicating a shift of funds into wealth management products.

Operating Changes in the First Nine Months of 2026

The third quarter report of October 30, 2026 turns the first half improvement narrative into a more complex picture. Revenue for the first nine months was 183 million renminbi, down 14.93 percent, reversing the 15.89 percent growth of the first half, and third quarter revenue alone fell 42.85 percent year over year, which is the main drag on the revenue line. Attributable net profit was 10.99 million renminbi, up 56.06 percent, and adjusted net profit turned positive at 1.00 million renminbi against a 3.25 million loss in the first half. Gross margin for the first nine months was 30.92 percent, up 5.11 percentage points, and the third quarter alone reached 39.61 percent, 13.42 points above the first half. Research spending was 35.29 million renminbi, 19.30 percent of revenue.

Three readings follow. On the revenue side, the near halving of third quarter revenue is consistent with the seasonality already identified in the first half, where second quarter revenue fell 52.0 percent sequentially, so this looks like contract size and acceptance timing weakening in the second half rather than a single-quarter anomaly. The statement that revenue maintained double-digit growth no longer holds at the nine-month level.

On the profit side, revenue fell while profit rose, and gross margin is the only explanatory variable. A third quarter margin of 39.61 percent sits more than fifteen points above the 24.77 percent full-year 2025 level, which usually signals a delivery mix shift toward higher-margin software or technical services, or improved project cost structure. Adjusted profit turning positive at 1.00 million renminbi is more informative than attributable net profit because it removes the effect of government subsidies and financial asset gains.

On durability, adjusted net profit of 1.00 million renminbi against 183 million renminbi of revenue is a 0.55 percent net margin, which is a marginal crossing into positive territory rather than established profitability. Research spending held at 19.30 percent of revenue with a broadly flat absolute amount, so research was not cut as revenue contracted. The real test is the fourth quarter: if the third quarter margin level carries into year end and the revenue decline narrows, the full year can be read as an inflection point.

Technology

Why Electromagnetic Simulation Is Hard, and Three Technical Routes

Electromagnetic simulation is difficult for three physical reasons. First, electrically large problems: when an antenna or platform is much larger than the wavelength, the degrees of freedom in a full-wave solve rise sharply and memory and time costs become prohibitive. Second, coexisting scales: a phased array contains millimetre-scale feed structures and metre-scale apertures, spanning three orders of magnitude in mesh size. Third, open boundaries: radiation and scattering require handling infinite space, and absorbing boundaries plus the singularities of integral equations add numerical difficulty.

Engineering practice has produced three routes. The finite element method suits complex inhomogeneous media and fine structures but is expensive at electrically large sizes. The method of moments, based on integral equations, discretizes surfaces only and suits radiation and scattering in open regions, but its matrices are dense and scale-limited. High-frequency asymptotic methods such as physical optics, geometrical optics and the uniform theory of diffraction are efficient at large sizes but depend on assumptions whose error is hard to control in complex structures. Commercial tools typically hybridize: HFSS is finite-element centric, CST is built on the finite integration technique with multiple solvers, and FEKO is method-of-moments centric with physical optics hybrids. RDSim follows the same hybrid family.

RDSim Technical Claims

RDSim is the company's general three-dimensional electromagnetic simulation product. The portfolio also includes specialized tools for antenna layout, antenna design optimization and phased array design optimization, plus application software for complex electromagnetic environment simulation and antenna fault diagnosis. The RDSim AI 2026 release of May 25, 2026 states three themes: AI surrogate models, engineering agents, and complex-scenario simulation.

The meaning of AI surrogate models in electromagnetics needs precision. It does not mean asking a language model to solve Maxwell's equations. It means training a neural network to approximate the mapping from design parameters to electromagnetic response, replacing full-wave solves inside optimization loops. The academic basis is the family of deep operator networks and Fourier neural operators: run a batch of high-fidelity simulations to build a training set, train a surrogate whose inference cost is negligible, use it for parameter sweeps and optimization, then verify key points with full-wave simulation. The engineering value sits in optimization, because array spacing, feed phase and dielectric thickness form a high-dimensional design space where each full-wave solve takes hours while a surrogate evaluates in milliseconds.

The three AI plus CAE themes disclosed in the 2025 annual report map onto this: intelligent electromagnetic inverse design, physics and data dual-driven simulation acceleration kernels, and large language model enablement across the simulation workflow. Physics and data dual-driven is another way of describing surrogate modeling, with the solver supplying training data and constraints and the data model supplying speed. Intelligent inverse design points at inverting the question from computing performance for a given structure to finding a structure for a given performance, which is the highest-value and hardest step in electromagnetic design.

Engineering Agents: Boundaries and Real Uses

Engineering agents are the other RDSim AI 2026 claim. Following industry practice, such agents handle orchestration: interpreting design requirements, invoking modeling tools to build geometry, setting materials and boundaries, submitting solves, reading results and generating reports. They do not replace solvers, nor do they replace engineering judgment.

This capability has real value in electromagnetic workflows because modeling time is often underestimated. Building a phased array model involves array layout, element geometry, feed networks and packaging, and manual modeling can take days, while parametric modeling with agent orchestration compresses that. The boundary is equally clear: electromagnetic correctness depends on physical understanding, and wrong boundary conditions, wrong port definitions or poor meshing produce plausible but incorrect results. Agents reduce operating time; they do not replace physical judgment.

Low-Altitude Physical AI as a New Entry Point

The 2025 annual report mentions a low-altitude physical AI simulation design and evaluation system developed for emerging demand. The technical substance is that communication links, navigation signals and radar detection for low-altitude vehicles such as drones and electric vertical take-off aircraft all depend on electromagnetic environment modeling, and low-altitude scenes feature complex terrain and building blockage, strong multipath and dense interference sources. Combining electromagnetic simulation, measurement verification and scene modeling is the technical basis for such systems.

Commercially the entry point merits attention because the low-altitude economy has explicit policy support and rigid demand for electromagnetic compatibility and spectrum safety. Public information does not disclose customers, contract values or delivery form for this system, so it remains a directional judgment.

The Simulation-to-Measurement Loop as the Most Substantive Differentiation

The most distinctive asset is owning both simulation software and measurement systems. That combination enables something pure software vendors cannot replicate: calibrating simulation models with measured data.

Electromagnetic simulation error has many sources. Dielectric constant and loss tangent vary with frequency and temperature, manufacturing tolerances shift geometry, feed networks carry parasitic parameters, and chamber measurements have systematic error. A model that converges cleanly on paper can deviate by several decibels in radiation pattern once built. The only reliable correction is measurement: capture pattern, gain, standing wave ratio and radar cross section in an anechoic chamber, then feed the difference back into the model by adjusting material parameters or introducing empirical corrections.

HollyWave's measurement systems cover antenna radiation, radar cross section, automotive communication, electromagnetic compatibility and automated radio frequency testing, which means it controls the input end of that feedback chain. In physical AI terms this is the embryo of a genuine data flywheel: measurement produces high-quality data that can train better surrogates, better surrogates raise simulation speed, and higher speed generates more design iterations and more measurement demand.

Whether the flywheel turns depends on two conditions: whether measurement data is structured and retained rather than archived with the report, and whether customers are willing to share data. The first is an internal capability and the second involves confidentiality. Public information does not describe the data platform, which makes this a diligence question.

Technical Shortfalls

First, solver coverage is narrow. HollyWave covers electromagnetics only, not structures, thermal or fluids. Real engineering problems are coupled: thermal distortion of a high-power antenna changes electromagnetic performance, and aerodynamic heating affects the dielectric properties of a radome. Such problems require coordination with other vendors' solvers, and no public multiphysics coupling solution exists.

Second, validation data for the AI capability is missing. RDSim AI 2026 has been available for about five months, and public information contains only feature descriptions, with no surrogate accuracy metrics relative to full-wave results, no speedup benchmarks by problem scale, and no quantified engineering cases. These are the key inputs for judging whether the AI capability is engineering-grade, and their absence limits assessment to direction.

Third, headcount constrains product breadth. Eighty researchers must maintain general electromagnetic simulation, specialized design tools, complex environment simulation, measurement system software and phased array products. Compared with the thousands of engineers at Ansys, Dassault and Keysight, depth of product refinement is necessarily constrained.

Product and Commercial Model

Commercial Roles of the Three Lines

Electromagnetic measurement systems are the cash engine, with 263 million renminbi of revenue in 2025 across phased array calibration measurement, radar cross section measurement, radio frequency measurement, 5G base station antenna over-the-air measurement, automotive millimetre-wave radar measurement, satellite measurement and hardware-in-the-loop simulation. The business is project-based: large individual contracts, long delivery cycles and revenue recognition tied to acceptance milestones, which explains violent quarterly swings. A 16.39 percent gross margin in 2025 indicates a material change in competition or cost structure.

Electromagnetic simulation and verification is the strategic direction, with 28.9 million renminbi in 2025, under one tenth of revenue. The portfolio includes the general RDSim product, specialized tools for antenna layout, antenna design optimization and phased array optimization, complex electromagnetic environment simulation software, and antenna fault diagnosis software. Software margins are typically much higher than equipment margins, but the scale is too small to offset the measurement margin decline.

Phased array products are an extension, with 0.66 million renminbi in 2025, a negligible scale. Their strategic role is to ground design and verification capability in delivered hardware and complete the design-verification-product chain, but they are not a revenue pillar in the near term.

A Major Adjustment to Funded Projects

On August 26, 2026 the company announced two things: termination of the funded industrialization project for intelligent testing systems for 5G large-scale antennas, and conversion of the digital phased array test and verification industrialization project into a new project for research and industrialization of a next-generation intelligent electromagnetic CAE software platform.

This is the key signal for understanding the strategic shift. Both moves point the same way: contracting test-related industrialization and redirecting raised funds toward a CAE software platform. There are two readings. The optimistic one is strategic focus, since measurement margins have fallen to 16.39 percent, competition in 5G large-scale antenna testing is intense and returns are poor, and software offers higher margins plus clearer policy support for domestic CAE. The cautious one is that the original projects missed their return assumptions, since changes to funded projects usually imply the original projections cannot be met.

The two readings are not mutually exclusive and may both hold. What matters for assessment is whether the new project delivers: how long a next-generation intelligent electromagnetic CAE platform takes to generate revenue, how many customers it can win, and how it is differentiated from the existing RDSim line. Announcements of this kind typically provide feasibility estimates without verifiable milestones.

Customers and Market

Customers concentrate in defense-adjacent fields including aviation, aerospace and shipbuilding, plus high-end manufacturing including commercial space, communications, automotive radar and the low-altitude economy. Such customers have rigid demand, long qualification cycles and explicit domestic sourcing requirements, but purchasing is sensitive to budget and acceptance timing. Notes and accounts receivable fell 40.29 percent in 2025, indicating better collection, yet operating cash flow in the first half of 2026 was negative 21.20 million renminbi, showing seasonal collection pressure persists.

On July 22, 2026 the company received a first prize for technological invention in the Shaanxi Provincial Science and Technology Awards for 2025. Such awards materially help qualification reviews with defense-adjacent customers, though the revenue contribution cannot be quantified.

Competitive Landscape

International Technical Barriers

The electromagnetic simulation market is relatively stable. Ansys HFSS is the industry benchmark for three-dimensional full-wave simulation and dominates antennas, radio frequency devices, packaging and electromagnetic compatibility. Dassault CST Studio Suite is strong in time domain and in wideband and transient problems. Altair FEKO is built on the method of moments and is strong in antenna placement on electrically large platforms. Keysight EMPro is tightly bound to its test instrument ecosystem.

The shared advantage of these four is three decades of solver accumulation, algorithm implementations validated across many engineering projects, and complete multiphysics and multi-domain coordination. Catching up in a single physics domain is feasible for domestic software, but replacing an entire toolchain is hard. RDSim overlaps functionally with HFSS and CST, and the gap lies mainly in solver robustness, convergence on complex structures and parallel efficiency at large scale, none of which can be judged from marketing language and all of which require head-to-head benchmark cases.

Domestic Competitors

A group of Chinese vendors has emerged with different emphases. Xpeedic entered from the EDA side with strength in package and system-level electromagnetics and signal integrity. Dongjun Technology positions a full-wave electromagnetic solver for electrically large and complex environment scenarios. Several university-affiliated teams build specialized tools. Most of these are pure software companies without measurement system businesses or phased array delivery capability.

HollyWave's relative advantage is therefore not a stronger solver but a shorter path to verification. When a customer needs simulation aligned with measurement, very few suppliers can provide software, anechoic chamber measurement and calibration services together. That combination helps in defense-adjacent projects and does not help in pure software procurement, where the customer may simply want a cheaper solver.

The Gap Between Substitution Logic and Reality

Policy favors domestic substitution: electromagnetic simulation is a priority area for industrial software autonomy, and defense-adjacent customers face procurement restrictions on foreign software, creating clear entry opportunities. The practical obstacle is migration cost. A design institute that has used HFSS for a decade has accumulated scripts, model libraries, material libraries and operating habits, and switching means rebuilding those assets and re-validating credibility.

The typical path for domestic electromagnetic software is therefore substitution in new scenarios first, with new projects, new frequency bands and new platforms adopting domestic tools while legacy projects keep existing ones. That path is slow but solid. HollyWave's measurement business plays a subtle role here: if a customer designs with RDSim and verifies with HollyWave measurement systems, the two data sets can be cross-checked, which strengthens the case for migration. That is a hidden advantage over pure software vendors.

Financing and Capital Structure

Star Market Listing and Raised Funds

The company listed on the Star Market in 2021 under code 688682, among the earlier cohort of Star Market companies with electromagnetic simulation as a core business. Proceeds were directed mainly at industrialization projects, including intelligent testing systems for 5G large-scale antennas and digital phased array test and verification systems. The August 2026 project change indicates the original direction needs adjustment, making the efficiency of raised fund use a metric to track.

The 2025 Profit Reversal

2025 revenue was 303 million renminbi, up 12.72 percent, with attributable net profit of negative 23.39 million renminbi against positive 14.43 million in 2024, and non-recurring-adjusted net profit of negative 29.85 million against positive 2.61 million. Basic earnings per share were negative 0.23 renminbi, weighted average return on net assets was negative 3.71 percent, down 5.92 percentage points, and return on invested capital was negative 3.75 percent.

Two direct causes were margin decline and higher research spending. Gross margin fell from 38.97 to 24.77 percent, a drop of 14.20 percentage points, with measurement systems falling from 34.70 to 16.39 percent, while research spending rose from about 42.19 million to 57.82 million renminbi, up 37.05 percent. With revenue growing only 12.72 percent, a 14 point margin decline plus 37 percent research growth is enough to push a company earning tens of millions into loss.

Operating cash flow deserves attention: positive 83.74 million renminbi in 2025, up 379.25 percent, in sharp contrast with the 23.39 million loss. The gap comes mainly from receivables collection, with notes and accounts receivable down 40.29 percent, plus inventory movements. Cash flow exceeding profit is a positive signal, and it also shows the profit deterioration came from margin and expenses rather than collection.

Valuation and Market Expectations in the First Half of 2026

At the August 25, 2026 close the trailing twelve month price to earnings ratio was about negative 168.37 times owing to losses, price to book about 5.47 times and price to sales about 10.2 times. At the 2025 annual report date of April 14, 2026 the price to sales ratio was about 15.56 times and price to book about 7.72 times. Multiples fell materially over four months, reflecting repricing of losses and growth quality.

A price to sales near ten times is not low for a company with 303 million renminbi of revenue, an operating loss and a revenue mix dominated by system integration. The support comes from the domestic electromagnetic CAE, commercial space and low-altitude economy narrative rather than current profit, which makes the share price highly sensitive to delivery. If the full year 2026 remains negative on an operating basis, valuation support shifts further toward thematic factors.

Financial Performance in the First Nine Months of 2026

The third quarter report of October 30, 2026 is the key document for understanding the company's current state. Revenue for the first nine months was 183 million renminbi, down 14.93 percent; attributable net profit was 10.99 million renminbi, up 56.06 percent; adjusted net profit was positive 1.00 million renminbi, up 171.16 percent, turning from negative to positive. Gross margin for the first nine months was 30.92 percent, up 5.11 percentage points. Third quarter revenue was 64.44 million renminbi, down 42.85 percent, with a single quarter gross margin of 39.61 percent. Research spending was 35.29 million renminbi, 19.30 percent of revenue.

Compared with the interim report, revenue flipped from 15.89 percent growth to a 14.93 percent decline, while profit flipped from reliance on non-recurring items to an adjusted profit that is positive. The two flips point in opposite directions and share one explanation: the company is contracting low-margin, low-value-added project delivery while raising the share of higher-margin business. That direction is consistent with the funded project pivot toward a CAE software platform, but the third quarter report does not disclose revenue by product, so this remains a directional inference rather than a verified explanation.

It should be kept in mind that adjusted net profit of 1.00 million renminbi is a 0.55 percent net margin on 183 million renminbi of revenue. Turning positive is a fact; established profitability is not. The valuation readings in this report remain on the pre-disclosure basis of the August 25, 2026 close, with price to earnings of about negative 168.37 times, price to book about 5.47 times and price to sales about 10.2 times, and require updating once post-report market data is added.

Dividends and Shareholder Returns

The 2025 distribution proposal was a cash dividend of 2.0 renminbi per ten shares before tax. Paying a dividend in a loss year indicates sufficient cash and retained earnings and a management preference for continuity in shareholder returns. Relative to a 23.39 million renminbi loss, the dividend is of similar magnitude. That choice is financially understandable, while as a signal it should be read together with the funded project change, since the company is redirecting money from test projects to a software platform while still paying cash.

DISCREPANCIES

1. The electromagnetic CAE positioning versus the actual revenue mix. The external narrative centers on RDSim electromagnetic simulation and media often label the company a domestic electromagnetic simulation software vendor. In 2025, however, simulation revenue was 28.9 million renminbi or 9.53 percent of revenue, while measurement systems were 86.67 percent. Both statements are accurate with different meanings: by revenue mix it is a measurement systems company, and by strategic investment and technology narrative it is a software company. This report uses a dual framing of measurement as cash engine and software as strategic direction rather than a single label. Source tier: OFFICIAL (annual report segment data) over MEDIA.

2. Attributable profit growth of 1,194.50 percent coexisting with an operating loss. First half 2026 attributable net profit was 4.38 million renminbi, up 1,194.50 percent, and media widely headlined profit growth of nearly twelve times. Non-recurring-adjusted net profit was a 3.25 million loss, with non-recurring items totaling 7.62 million including 6.50 million in government subsidies. Both figures are real, but the comparison base was extremely low at about 0.34 million renminbi and the profit source is non-operating. This report lists attributable profit, adjusted profit and the composition of non-recurring items together. Source tier: OFFICIAL (interim report).

3. Revenue growth and the swing to loss reported side by side in 2025. Many reports headline revenue growth of 12.72 percent with the loss in the body. The substantive change in operating quality is margin: overall gross margin fell 14.20 percentage points and measurement systems fell 18.31 points. Growth and deteriorating profit occurred simultaneously, indicating growth was bought with price or cost structure. This report presents the margin change as a core fact rather than citing revenue growth alone. Source tier: OFFICIAL.

4. Half-year and single-quarter margin figures are easily conflated. First half 2026 gross margin was 26.19 percent, down 3.7 points, while second quarter gross margin was 26.72 percent, down 4.3 points year over year but up 0.79 points sequentially. Commentary on whether margin has stabilized often mixes the two periods and reaches opposite conclusions. This report labels the period whenever either figure appears. Source tier: OFFICIAL.

5. Second quarter revenue growth of 32.16 percent versus a sequential decline of 52.0 percent. Quarterly revenue was 38.40 million renminbi, up 32.16 percent year over year and down 52.0 percent sequentially. Reading only the year-on-year figure suggests acceleration and reading only the sequential figure suggests collapse, while the actual cause is seasonality in delivery and acceptance tied to customer budget cycles. This report gives both rates and explains the cause. Source tier: OFFICIAL.

6. Inconsistent wording on the RDSim AI 2026 release date. The corporate website news item is dated May 25, 2026, while the interim report states the release happened during the reporting period without a specific date, and secondary coverage therefore writes first half of 2026 or attributes it to the second quarter. All three are valid with different precision. This report uses the website date of May 25, 2026 and notes that the interim report gives no specific date. Source tier: OFFICIAL (website) and OFFICIAL (interim report) in parallel, with the website date preferred.

7. AI surrogate models and large language model enablement are conflated. The company describes two distinct technical lines: AI surrogate models, which are reduced-order models used to accelerate optimization in place of full-wave solves, and large language model enablement across the simulation workflow, which handles orchestration and interaction. The first is numerical acceleration and the second is interaction automation, and their technical natures differ completely. Some coverage calls both AI simulation. This report separates them in the technology section. Source tier: OFFICIAL (annual report technical description) plus industry technical knowledge.

8. Two readings of the funded project change. The August 26, 2026 announcement terminated the intelligent testing system industrialization project for 5G large-scale antennas and converted the digital phased array test and verification project into a next-generation intelligent electromagnetic CAE software platform project. The company framing is focus and better use of funds; an alternative reading is that the original projects fell short of return expectations. Both require later data on the new project's progress, customer wins and revenue contribution. This report presents both readings without a one-sided conclusion. Source tier: OFFICIAL (announcement) plus inference (this report's analysis).

9. Positive first-half revenue growth coexisting with a negative nine-month figure. The interim report records revenue of 118 million renminbi, up 15.89 percent, while the third quarter report records first nine months revenue of 183 million renminbi, down 14.93 percent, with the third quarter alone down 42.85 percent. Both come from periodic reports and both are true, but they cover different periods, and citing only the interim figure overstates the current revenue trend. The profit direction also flipped: the first half still carried an adjusted loss of 3.25 million renminbi, while the first nine months turned positive at 1.00 million renminbi. This report uses the third quarter report as the current basis and retains the interim figures as half-year facts. Source tier: OFFICIAL (interim report) and OFFICIAL (third quarter report) in parallel, with the third quarter report as the current basis.

Risks and Diligence Gaps

First, whether measurement system margin can recover. Measurement gross margin was 16.39 percent in 2025 and overall gross margin in the first half of 2026 was 26.19 percent, down 3.7 points year over year. The third quarter report gives a positive signal: gross margin for the first nine months was 30.92 percent, up 5.11 points, and the third quarter alone was 39.61 percent. Whether the recovery holds is still open, because revenue fell 14.93 percent over the same period and the third quarter alone fell 42.85 percent, so the margin improvement may reflect a delivery mix shift rather than improved unit economics. Unless margin holds above 30 percent, adjusted profit cannot reach a meaningful scale even if revenue stabilizes. Segment gross margins would be needed to verify this.

Second, the software business is too small to offset hardware volatility. Simulation revenue of 28.9 million renminbi or 9.53 percent of the total, even growing above 30 percent, adds under 10 million renminbi annually, which cannot offset tens of millions of renminbi in profit impact from measurement margin decline. In the near and medium term the company remains structured so that hardware cycles determine results.

Third, validation data for the AI features is absent. Accuracy of the RDSim AI 2026 surrogate, speedup ratios and applicable problem scales are all undisclosed. Without them it is impossible to judge whether the AI capability is engineering-grade or demonstration-grade, which is the most important gap in assessing physical AI content.

Fourth, customer concentration and budget cycles. Customers concentrate in defense-adjacent and aerospace fields where purchasing depends on annual budgets and acceptance milestones, producing extreme quarterly volatility, with the second quarter of 2026 down 52.0 percent sequentially. Payment terms are long, and first half 2026 operating cash flow was negative 21.20 million renminbi. The absolute size and aging of receivables require tracking.

Fifth, valuation dependence on themes. With price to sales near 10.2 times and an operating loss, valuation support comes mainly from domestic substitution, commercial space and the low-altitude economy. If policy pacing or order conversion in any of the three falls short, downside room is significant.

Sixth, execution risk after the funded project change. Building a next-generation intelligent electromagnetic CAE platform takes time and requires differentiation from the existing RDSim line. If the new product overlaps, internal resources may fragment; if it is positioned too high, it may drift from existing customer needs.

Seventh, constraints from headcount. Eighty researchers covering three product lines is a heavy per-person load. Electromagnetic simulation talent is scarce, and recruiting and retention costs are high at average research compensation of 426,200 renminbi. Loss of core technical staff would directly affect product iteration.

Composite Assessment

Technology scores 7.5. The combination of electromagnetic simulation and measurement verification is scarce in China, the simulation-to-measurement data loop is a genuine technical asset, the RDSim surrogate direction aligns with industry trends, and the low-altitude physical AI entry point is well chosen. Deductions for single-domain solver coverage, absent quantification of AI features, and constrained product refinement owing to research scale.

Business scores 6.5. Revenue of 303 million renminbi is growing at a double-digit rate, measurement systems provide a stable cash source, and customers sit in high-value fields. Deductions because the mix depends heavily on system integration at 86.67 percent, software is under one tenth, and margin has fallen sharply with no clear recovery path.

Finance scores 6.0. Low leverage, improving cash flow and 83.74 million renminbi of operating cash flow in 2025 are strengths. The third quarter report shows gross margin for the first nine months up 5.11 points to 30.92 percent and adjusted net profit turning positive at 1.00 million renminbi, so earnings quality is improving. Deductions for the 2025 swing to loss with adjusted loss of 29.85 million renminbi, a 14.93 percent revenue decline over the first nine months with the third quarter alone down 42.85 percent, an adjusted net margin of only 0.55 percent that amounts to a marginal crossing into positive territory, continued reliance on non-recurring items, and a dividend paid in a loss year.

Product scores 7.0. Three lines form a design-verification-delivery loop, the three RDSim AI 2026 capabilities are directionally right, and specialized tools for antenna layout, phased array optimization and fault diagnosis address specific needs. Deductions for missing verifiable AI metrics, high functional overlap with international vendors, and a funded project change showing the product roadmap is still adjusting.

Market scores 7.5. Policy support for domestic electromagnetic CAE is explicit, defense-adjacent access creates certain demand, commercial space and the low-altitude economy add incremental scenarios, and the Shaanxi technology award strengthens qualification credentials. Deductions because substitution is constrained by migration cost, domestic competitors are multiplying, and valuation already prices in considerable thematic expectation.

Overall 6.9. This is a company with a scarce technology combination and unproven commercial conversion: it holds a rare domestic capability spanning electromagnetic simulation and measurement verification and is genuinely investing in physical AI through surrogate models, engineering agents and low-altitude simulation, yet software is only 9.53 percent of revenue, AI has no separate pricing, and adjusted profit has been negative. The judgment on physical AI is that the physical core of electromagnetic solving plus measured calibration is real while the AI layer is early, and that investment value depends on margin recovery in measurement systems and scale in software, both requiring two to three quarters of data.

References

  1. HollyWave 2025 annual report, disclosed April 15, 2026: revenue 303 million renminbi up 12.72 percent, attributable net profit negative 23.39 million renminbi versus positive 14.43 million, adjusted net profit negative 29.85 million versus positive 2.61 million, operating cash flow 83.74 million versus negative 29.99 million, basic earnings per share negative 0.23 renminbi, weighted average return on net assets negative 3.71 percent down 5.92 points, return on invested capital negative 3.75 percent, gross margin 24.77 percent down 14.20 points, dividend proposal 2.0 renminbi per ten shares
  2. HollyWave 2025 annual report segment data: electromagnetic measurement systems 263 million renminbi at 86.67 percent of revenue, up 11.99 percent, gross margin 16.39 percent down 18.31 points; electromagnetic field simulation and verification 28.9 million renminbi at 9.53 percent; phased array products 0.66 million renminbi at 0.22 percent; other 10.86 million renminbi at 3.58 percent; all revenue domestic
  3. HollyWave 2025 annual report research data: research spending 57.82 million renminbi up 37.05 percent and 19.07 percent of revenue, five-year compound growth 13.32 percent; 80 research staff versus 77, 46.24 percent of employees versus 44.25 percent; average research compensation 426,200 renminbi including share-based payment
  4. HollyWave 2025 annual report technology disclosure: anchoring a long-term AI plus CAE technical line, with work on intelligent electromagnetic inverse design, physics and data dual-driven simulation acceleration kernels, and large language model enablement across the workflow; full iteration of RDSim and complex electromagnetic environment simulation software; release of a low-altitude physical AI simulation design and evaluation system; customers covering electronic information, aviation, aerospace, shipbuilding, commercial space and the low-altitude economy
  5. HollyWave 2026 interim report, disclosed August 26, 2026: revenue 118 million renminbi up 15.89 percent, attributable net profit 4.38 million up 1,194.50 percent, adjusted net profit negative 3.25 million versus negative 4.28 million, operating cash flow negative 21.20 million versus negative 48.64 million, basic earnings per share 0.043 renminbi, weighted average return on net assets 0.71 percent, research at 19.45 percent of revenue, gross margin 26.19 percent down 3.7 points, net margin 2.36 percent
  6. HollyWave 2026 interim report non-recurring items: total 7.62 million renminbi, including 6.50 million in government subsidies, 2.57 million in fair value changes and disposal gains on financial assets, and negative 1.37 million in tax effects
  7. HollyWave second quarter 2026 data: revenue 38.40 million renminbi up 32.16 percent year over year and down 52.0 percent sequentially, attributable net profit negative 2.11 million up 44.51 percent year over year and down 132.54 percent sequentially, adjusted net profit negative 8.29 million down 28.77 percent year over year and down 264.32 percent sequentially, gross margin 26.72 percent down 4.3 points year over year and up 0.79 points sequentially, net margin negative 8.32 percent; total assets 864 million renminbi down 8.9 percent from the prior year end and attributable net assets 595 million down 2.6 percent
  8. HollyWave announcement of August 26, 2026 on funded projects: termination of the intelligent testing system industrialization project for 5G large-scale antennas; conversion of the digital phased array test and verification industrialization project into a next-generation intelligent electromagnetic CAE software platform research and industrialization project
  9. HollyWave website news, New Version Release, RDSim AI 2026, from Efficient Solving to AI-Assisted Analysis, dated May 25, 2026: RDSim AI 2026 three-dimensional electromagnetic simulation software formally released, positioned as a shift from tool-driven to intelligence-driven, emphasizing the balance of high-precision solving, efficient iteration, complex modeling and automated delivery
  10. HollyWave website product and business pages: electromagnetic CAE software covering modeling, solving, optimization and effectiveness assessment; measurement systems covering antenna radiation, radar cross section, automotive communication, electromagnetic compatibility and automated radio frequency testing; phased array products covering array design, beam control and lightweight integration
  11. Valuation data: at the August 25, 2026 close, trailing twelve month price to earnings about negative 168.37 times, price to book about 5.47 times and price to sales about 10.2 times; at the April 14, 2026 close, trailing twelve month price to earnings about negative 201.78 times, price to book about 7.72 times and price to sales about 15.56 times
  12. HollyWave website news of July 28, 2026: on July 22, 2026 the Shaanxi provincial authorities announced the 2025 science and technology awards, and Shanghai HollyWave Electronic System Technology received a first prize for technological invention
  13. HollyWave 2026 third quarter report, disclosed October 30, 2026: revenue for the first nine months 183 million renminbi down 14.93 percent, attributable net profit 10.99 million up 56.06 percent, adjusted net profit 1.00 million up 171.16 percent turning from negative to positive, third quarter revenue 64.44 million down 42.85 percent, gross margin for the first nine months 30.92 percent up 5.11 points, third quarter gross margin 39.61 percent, research spending 35.29 million renminbi at 19.30 percent of revenue
This report is compiled from public sources for informational purposes only and is not investment advice. Financial figures follow company periodic reports and announcements, competitive and technical judgments are analytical, and valuation data change with market prices.

Related tools

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.