Data for AI is being built into a recurring business anchored in AI retraining demand. AI-use corporate licenses are driving growth, and 2H hinges on executing the order backlog.

Summary

1H cumulative revenue was flat YoY. AI-use corporate licenses lifted License-Based revenue. However, Project-Based revenue declined on a tough comparison against a new North American data development project in the prior-year period, and the adjusted EBITDA loss widened. Management stressed that retraining AI models requires reuse of historical data. It therefore plans to develop Data for AI as a recurring license business rather than a one-off sale. Operating CF and FCF turned positive, and full-year guidance was maintained. Real-world deployment of 3D data also advanced, with three MLIT logistics automation projects and a new port project selected.

Key Points (Earnings Takeaways and Growth Initiatives)

• Management Strategy and Market View
  • AI training outputs weights, but performance gains require ongoing retraining. Retraining on new data alone causes bias and forgetting, so historical training data must also be used.
  • Built on this retraining cycle (data flywheel), the company envisions recurring contracts rather than one-off data provision.
  • The company positions itself as a provider of the data and systems that let AI understand the real world, not as an AI developer. Management believes data demand will persist as E2E models evolve, driven by regional differences and expanding use cases.
  • The operating environment is largely unchanged from 1Q. Management sees growing Physical AI investment as a tailwind but continues to monitor uncertainty in the auto industry.
• Current Business Progress and Drivers
  • License-Based revenue growth was not driven by any single deal. Three factors contributed: expanded regional coverage for existing customers, new customers in new industries, and more vehicle models adopting the data in mass production.
  • A major overseas semiconductor manufacturer placed follow-on orders for European, Korean and Japanese data after using North American data. Region-specific road environments and traffic rules are what generate data demand.
  • The improvement in operating CF was driven mainly by collection of trade receivables. Higher contract liabilities from expanding mass-production licenses also contributed.
  • 3D data licenses are gaining traction in real estate development and infrastructure management. Management acknowledges, however, that a meaningful revenue contribution requires further expansion of customers and use cases.
• Key Strategic Initiatives and Inflection Points
  • The company's 3DGS combines imagery with point clouds and high-definition 3D maps. Its differentiator is the ability to use generative AI to alter weather, time of day and traffic participants while preserving geometric structure.
  • Real-world driving data and 3DGS-generated data were fed into a public E2E model and the outputs compared. Effectiveness was validated on model behavior, not visual appearance.
  • AI-native data was recently delivered to the R&D division of a major technology company, broadening the customer base into Physical AI research.
  • A company proposal was newly selected under MLIT's Port Technology Development Program. The project will build an information-sharing platform for vehicles and cargo-handling equipment within ports, with details to be announced once ready.

Outlook and Strategy

• Full-year guidance was maintained at revenue of JPY 7,000M and adjusted EBITDA of JPY 50M. 1H actuals plus order backlog total JPY 6.2B. The remaining approx. JPY 800M must be converted from a JPY 1.34B pipeline (our estimate).
• 1H revenue progress stands at 35.1% of the full-year plan (our estimate). Management explained that customer acceptances and corporate license bookings are structurally concentrated in the second half. It noted that JPY 2.3B was booked in 4Q alone last fiscal year.
• The medium- to long-term revenue growth target is 10–30% annually. Some AI-use corporate license deals carry 100% gross margins, and the company targets a marginal profit ratio above 80% for the License-Based business overall.
• The three logistics demonstration projects are being developed as in-house systems, not one-off pilots. The plan is to standardize know-how, roll it out horizontally to warehouses, ports, airports and factories, and license it.
• Management noted that corporate licenses for autonomous trucks could transition to mass-production licenses as mass production and commercialization progress.
• On M&A, the company continues to consider a roll-up of surveying firms and acquisitions in downstream domains. Group synergies are materializing, including new wins from drone surveying and insourcing of outsourced work.

Positive Factors

• The corporate license customer base has broadened from semiconductor manufacturers to autonomous driving system developers, autonomous truck developers and major technology companies.
• Through semiconductor manufacturers, the data reaches downstream automakers and developers in an ecosystem structure. Management intends to eventually supply these data users directly, reducing dependence on specific customers.
• In mass-production licenses, the data was newly adopted in the Nissan Elgrand, and Automaker B began selling a new model equipped with the company's data.
• The company was selected for MLIT projects in three areas: intra-facility transfer transport at logistics facilities, mobility hubs and automated logistics roads. Management believes the commercialization market for autonomous trucks has taken off earlier in North America than in Japan.
• On October 28, a consolidated subsidiary disclosed an order for road data development in Saudi Arabia (approx. JPY 670M).
• On October 27, the company announced it is exploring a collaboration with Toshiba on an infrastructure-cooperative mobility platform.
• Contracts with municipalities for the snow-removal support system are progressing well ahead of this winter. 3Dmapspocket has been adopted by Tokyo Tatemono, SYLA and Mitsui Fudosan Residential.

Concerns and Risks

• The company posted a 1H gross loss of JPY -419M. Cost of sales rose by JPY 633M, including a JPY 156M increase in depreciation. Improving profitability depends on building up high-margin license revenue in 2H.
• Achieving full-year guidance is 2H-weighted. Management itself acknowledged the risk that even booked orders could slip depending on customers' development schedules and acceptance timing.
• The customer breakdown of corporate licenses is not disclosed. This makes it difficult for outsiders to gauge the impact if a major customer were to reduce or change its contract.
• 3DGS remains at the technical validation stage using public data and public models. Management stated it does not consider the output fully equivalent to real-world driving data. Building a track record of effectiveness in live projects using customer data is a key item to monitor.
• Net assets declined from JPY 7,229M at end-March 2026 to JPY 5,688M. Long-term borrowings increased to JPY 1,424M, and a JPY 2,000M commitment line has been secured.
• External uncertainties remain. These include weaker auto industry earnings from geopolitical risk and higher oil prices, the local situation in the Middle East, and map data re-creation costs related to the Kumamoto earthquake (extraordinary loss of JPY 32M).

Performance Highlights

For 1H FY3/27, License-Based growth offset the Project-Based decline, leaving revenue flat YoY. Adjusted EBITDA was JPY -1,063M, a JPY 497M deterioration YoY. Interim Net Loss Attributable to Owners of Parent Company was JPY 1,723M, while operating CF turned positive at JPY 347M.

Segment Performance

*Growth rates other than License-Based are our estimates. Operating income is disclosed on a consolidated basis only. As the prior-year period was also a loss (JPY -1,142M), no growth rate is shown.

• License-Based Revenue Share: 52.2% (+13.7pt YoY, our estimate)
• Overseas Revenue Share: 86.5% (+12.9pt YoY, our estimate)
• Free Cash Flow: JPY 27M (YoY ―)
• Cash and Deposits: JPY 3,813M (+JPY 155M vs. end-March 2026)
• Adoption in Mass-Production Vehicles: Cumulative 6 automakers / 39 models (YoY ―)
• Cumulative AI-Native Data Downloads: Over 4,500 (YoY ―)

Q&A List

• Q

  I understand delivery of the Middle East map development has been completed. Has the corresponding revenue been fully booked through 2Q? If not, when will it be recognized?

  A

  We understand this question refers to the new high-definition 3D data development project we recently disclosed. The project as a whole is still in progress, and delivery has not been completed. Revenue is not recognized in full upon delivery. It is recognized according to project progress, and the remaining revenue will likewise be booked as the project advances.

• Q

  The 35.8% YoY increase in License-Based revenue is very encouraging. Which customer segments or use cases drove the growth?

  A

  One growth driver was the increase in automotive corporate licenses, centered on AI use cases. First, we expanded regional coverage for existing customers. Specifically, after we supplied North American data to a major overseas semiconductor manufacturer, the customer requested European, Korean and Japanese data as well, and we won follow-on orders. This deal directly translated into revenue by showing we can deploy our global high-definition 3D data to the same customer across multiple regions and use cases. Our customer base is also widening. Beyond semiconductors, we are increasingly supplying autonomous driving system developers, autonomous truck companies and major technology companies with data for Physical AI training, evaluation, validation and simulation — what we call Data for AI. On top of conventional 3D data for mass-production vehicles, this expansion into AI use cases is driving growth in corporate licenses rather than mass-production licenses. Mass-production licenses are also being adopted in new models, bringing cumulative adoption to 6 automakers and 39 models. This license revenue growth was not driven by any single deal. It reflects the combined contribution of expanded regions and use cases for existing customers, expansion into new industries and customers, and more vehicle models adopting our data.

• Q

  What initiatives will you take to grow License-Based revenue further?

  A

  We hold data globally. We aim to leverage these data assets to broaden the scope of supply to existing customers while developing and proposing new use cases to new customers, steadily building License-Based revenue.

• Q

  Operating cash flow and free cash flow have turned positive. Please explain the main drivers, such as collection of receivables, contract liabilities and capex.

  A

  The main driver of the improvement in operating cash flow was collection of trade receivables. In addition, contract liabilities increased with the expansion of mass-production licenses. Contract liabilities are the portion of consideration received under mass-production license agreements that will be recognized as revenue over the remaining contract term. In 1H, the increase in contract liabilities also contributed to the improvement in operating cash flow. Meanwhile, we continue to make growth investments, including data generation in new countries and regions — our 27th and 28th countries — and M&A. Despite these investments, free cash flow (operating cash flow plus investing cash flow) turned positive in 1H.

• Q

  Please explain your capital allocation policy going forward.

  A

  We will continue investing in building datasets that lead to future License-Based revenue, in products, and in M&A. Management will make capital allocation decisions while assessing investment priorities, profitability and capital efficiency.

• Q

  You maintained full-year guidance despite a 1H loss, implying substantial 2H revenue and profit. What revenue do you expect to book in 2H to meet the full-year target, including the progress of booked orders and assumptions for additional orders?

  A

  We maintained full-year guidance based on 1H results and the current order situation. Our plan is to execute booked orders scheduled for 2H revenue recognition and to win additional deals. In Project-Based, we are progressing automotive data development projects in the Middle East and a new region. We will also execute domestic government projects and the three logistics automation orders introduced today, recognizing revenue according to progress. In License-Based, we expect to build revenue from three sources. These are corporate licenses mainly for AI training, evaluation and validation; mass-production licenses on the back of more equipped models; and 3D data licenses for infrastructure management and real estate development. 3D data licenses are small in scale. Our results tend to concentrate project acceptances and corporate license revenue in the second half, as past fiscal years show. Indeed, last fiscal year we booked JPY 2.3B in 4Q, mainly from corporate licenses. Our business structure means 1H results alone do not fully reflect full-year progress. Through these revenue bookings, combined with cost control and efficient use of group resources, we aim to achieve full-year revenue of JPY 7.0B and adjusted EBITDA of JPY 50M. Based on current project progress, we see no need to revise guidance and have maintained it.

• Q

  Regarding the deals underpinning full-year guidance, please break down the revenue composition and certainty, to the extent possible, into 1H actuals, booked orders and pipeline under negotiation.

  A

  The path to full-year guidance is broken down into three tiers — actuals, booked orders and pipeline — on page 40 of the earnings presentation. The first tier is 1H actuals, i.e., revenue recognized through 2Q. The second tier is booked orders scheduled for revenue recognition in 2H. Partly overlapping with the previous answer, in Project-Based we have booked domestic and overseas data development projects and logistics automation projects that we will now execute. In License-Based, mass-production licenses are booked. Some AI-focused corporate licenses are also booked but not yet delivered, and we are executing these toward revenue recognition. The third tier is pipeline under negotiation. These are deals where customer discussions have advanced and a certain level of certainty is expected, and we are working to win contracts and recognize revenue. Guidance will be achieved by adding booked orders to 1H actuals and converting pipeline deals into contracts and revenue. However, even for booked orders, there is a risk that revenue timing may shift depending on customers' development schedules and acceptance timing. Recognizing these risks, we are closely managing the progress of each deal to achieve full-year guidance.

• Q

  I see Data for AI as a major growth opportunity that allows DMP's data assets to be monetized repeatedly. How will you develop it into recurring license revenue by adding regions and use cases?

  A

  There are two basic monetization models. The first provides imagery, point clouds and high-definition 3D data developed worldwide as corporate licenses for AI-native uses such as training, evaluation, validation and simulation. The second combines customers' existing imagery and driving data with our data to generate 3D environment models and evaluation scenarios tailored to customers' development goals — essentially turning a flywheel. The two are not separate but deployed as a package. When customers actually use our data in development, new regions, use cases and data needs become clear. This leads to additional data supply, further generation and processing, and updates, as we aim to turn the data flywheel together with customers. In practice, we initially supplied North American data to a major overseas semiconductor manufacturer, and building on that, we are expanding coverage to Europe, Korea and Japan this fiscal year. In AI-based autonomous driving and ADAS development, road environments and traffic rules differ by region. Real-world data matching each service region is therefore required. Even if a vehicle can drive in a region, data reflecting regional differences is needed to assure safety there. Revenue opportunities go beyond adding regions. As development progresses, use cases can extend from training to validation simulation, safety verification, regulatory certification, and inference rather than training. This could unfold along the lines of the five Data for AI categories we have previously explained. We therefore see Data for AI not as a simple one-off data sale but as a license business that builds through added regions and use cases, continued usage periods, data updates and new customer wins. While we cannot guarantee individual follow-on orders or renewals, we will continue building a subscription-type, recurring model for corporate licenses.

• Q

  Even if 3DGS-generated data looks realistic, it may not hold the same value for AI as real-world driving data. Please clarify what practical utility has been confirmed so far and what has not.

  A

  As you point out, our technical validation is not based on visual realism alone. As shown in the demo video, we fed each environment model into a public E2E model to confirm usability, i.e., whether it operates properly, and presented data comparing the outputs. From a single 3D environment model, we can generate data matched to different camera positions, fields of view and driving trajectories. We can also freely add traffic participants such as oncoming vehicles and reproduce a variety of scenarios. This allows us to verify whether the E2E model operates correctly. However, we are currently at the technical validation stage using public data and public models. We do not consider the output fully equivalent to real-world driving data. Going forward, we will support AI model training, evaluation and validation by combining 3D environment models generated from data held by automakers and other customers with our high-definition 3D data and AI datasets.

• Q

  How do you assess the risk that advances in autonomous driving AI, including E2E models, reduce reliance on HD maps and structured data? Why will DMP's data remain necessary for AI training, evaluation, validation and safety verification?

  A

  As a basic premise, we are not a company that develops AI itself. We provide the data and systems that allow AI to accurately understand the real world and to be trained, evaluated and validated. Generative AI becomes smarter by automatically learning from text and images online. Physical AI in autonomous driving, construction machinery, industrial machinery and robotics likewise learns by understanding the real world. I would like to explain how data continues to be used there. Having already covered the need for additional data by region and use case, I will explain from a more technical angle. Data does not become unnecessary once an AI model has been trained. To use a cooking analogy, the AI model is the recipe procedure — putting water in a pot and heating it. Data is the ingredients and the accumulated cooking experience. Weights, which always come up in AI development, quantify through repeated training how much each factor matters, such as whether A or B is more influential. They are a massive block of numerical information. In cooking terms, they correspond to ingredient weights in grams, pot temperature and heat level, which let you reproduce the dish exactly. But that only gets you one dish. To make the dish better — that is, to advance AI model training — you improve by iterating with new ingredients as well as past ingredients and cooking experience. Likewise in AI, retraining to improve performance also requires the original training dataset, so data must be supplied continuously. Training on new data alone causes forgetting of what was learned before, just as cooking only with new ingredients produces an unbalanced dish. Training must therefore continue while blending in past data. In other words, once model training begins, customers keep using our data alongside the weights throughout that period. We envision building this into a data business with recurring characteristics.

• Q

  Corporate license growth appears heavily dependent on a few large customers. Wouldn't revenue be significantly affected if such a customer scaled back its contract or changed its strategy? Please explain the risks and countermeasures.

  A

  We do not disclose revenue composition or contract terms by individual customer. Top customers for the combined Project and License businesses are disclosed in our Annual Securities Report, but the corporate license breakdown is not. We aim to diversify corporate licenses as much as possible across three axes: customers, use cases and regions. On customers, the major overseas semiconductor manufacturer naturally carries significant weight. However, our base is broadening to automakers, autonomous driving system developers, autonomous truck developers and other technology companies. Downstream of semiconductor manufacturers are automakers and autonomous driving system developers, the end customers adopting GPUs and other products, systems and software development tools. So even when our direct customer is a semiconductor manufacturer, our data is used by a wide range of downstream users and development projects. We believe this leads to deployment across the entire autonomous driving, ADAS and Physical AI ecosystem. Going forward, we also plan to supply these data users directly to further broaden our customer base. On use cases, as explained in our five categories, applications range from AI training and AI validation to use as complementary or supplementary information for AI and regulatory approval. Regional coverage is also expanding from North America to Europe, Korea and Japan. We will continue to broaden customers, use cases and regions through both ecosystem-driven deployment and direct customer development, reducing dependence on specific customers.

• Q

  I see significant potential in horizontal expansion into logistics facilities, warehouses, ports and factories. How will you turn current demonstration data, technology and know-how into common products and recurring license revenue?

  A

  In logistics automation, we already supply high-definition 3D data to multiple customers developing autonomous trucks in Japan and North America, monetized as corporate licenses. The autonomous trucks under development appear close to mass production and commercialization. These could transition and evolve into mass-production licenses based on the number of units equipped. In addition, we are advancing projects in the areas explained today: intra-facility transfer transport at logistics facilities, receiving systems for autonomous trucks, and converting public roads into automated logistics roads. In each project, beyond our high-definition 3D data and information-sharing platform, we integrate with third-party systems such as berth management and facility management systems. Together with all stakeholders, we aim to provide logistics automation as a common platform. For now, we will execute these as Project-Based revenue. We then intend to commonize and standardize the resulting technology and operational know-how and roll it out to logistics facilities, warehouses, ports, airports and factories, leading to licensing of data products and systems. We do not view these as one-off pilots at all. We are developing them as our own systems and envision a business model in which each site uses them on a license basis. On the market environment, we sense firsthand that the commercialization market for autonomous trucks has taken off further in North America than in Japan. In Japan, in addition to the three projects we won, we hope to soon release a project to enhance port safety. With government initiatives also advancing, we expect the market to expand going forward, as in the US.