CADDi Series D More Than Doubles Its Valuation, but Factory Data Is the Real Test

CADDi raised $114 million in Series D funding at a $1.2 billion valuation, more than doubling the manufacturing AI startup’s reported value since March 2025. The CADDi Series D turns a specialist software company into a closely watched test of whether AI can work with the drawings, production records, and undocumented decisions inside factories.

The Tokyo and Chicago-based company is not promising a better general-purpose chatbot. It wants to build a shared data layer across computer-aided design files, technical drawings, purchasing records, enterprise software, and other manufacturing systems. Its software then uses that connected information to support search, design reviews, purchasing decisions, and specialized AI agents.

That approach puts CADDi against a more entrenched opponent than any single startup. Manufacturers already rely on product lifecycle management, enterprise resource planning, document storage, and engineering tools from established vendors. CADDi must prove that a new intelligence layer can connect those systems without becoming another isolated repository.

The CADDi Series D Reprices a Manufacturing AI Bet

The new financing values CADDi’s data strategy, not just its ability to search technical drawings.

The round was reported on September 15, 2026. Eight new and existing investors participated, according to the company and a funding account published by Fortune.

New investors included Moore Strategic Ventures, Coreline Ventures, Salesforce Ventures, Toyota-backed Woven Capital, and Recruit Holdings’ HR Tech Fund. Existing investors Atomico, Globis Capital Partners, and funds managed by a Japan Post Bank subsidiary also participated.

CADDi’s Japanese Series D announcement described the transaction as ¥17.7 billion and placed its corporate value at ¥182 billion. It said cumulative equity funding had reached ¥36.31 billion.

The financing follows a major repricing in a short period. CADDi reported a $470 million valuation in March 2025. The latest $1.2 billion figure is about 2.6 times that amount.

The company says the round brings its total funding to $234 million. That capital will support four priorities: proprietary AI development, expansion of the manufacturing data platform, international growth centered on North America, and hiring.

Those priorities show how CADDi’s ambition has changed. The company began with a narrower problem involving duplicate parts, technical drawings, and procurement decisions. It now wants to become the information foundation beneath multiple engineering and production workflows.

The original product, CADDi Drawer, ingested technical drawings and found similar parts in a manufacturer’s records. That search could help a customer reuse inventory, return to an existing supplier, or reconsider a purchasing decision.

CADDi has since renamed that product CADDi Explorer and surrounded it with a larger software portfolio. CADDi Agent handles defined manufacturing analyses, while workflow applications target design review, cost analysis, quotations, and production readiness.

This expansion matters more than the unicorn label. Search is a useful entry point because employees immediately recognize the frustration of hunting for drawings. The larger commercial opportunity depends on turning that search index into infrastructure used throughout a product’s life cycle.

CADDi says sales are more than doubling annually, although it has not disclosed revenue or customer totals. It reports customers in 22 countries and says more than half of Japan’s 100 largest manufacturers use its software.

Those are company-provided figures rather than audited public-company metrics. Still, they help explain why investors are backing an expansion rather than a single-feature product.

The round also gives CADDi time to pursue enterprise deployments, which usually involve security reviews, integrations, data preparation, and employee training. These projects cannot be evaluated using consumer software adoption cycles.

The financing therefore creates a demanding benchmark. CADDi must show that its widened product line produces deeper, repeatable use across engineering, procurement, quality, and production teams.

Why Manufacturing Data Became the Prize

Manufacturers do not lack information; they lack a reliable way to connect information created by different people, formats, and systems.

A technical drawing carries more than geometry. It can contain materials, tolerances, surface treatments, dimensions, revision histories, and manufacturing notes. Related decisions may live in purchasing systems, quality reports, spreadsheets, email, or an experienced engineer’s memory.

General-purpose language models work best with accessible text. They are less dependable when the task requires interpreting geometry, matching visually similar components, or tracing a design change through production and quality records.

CADDi says it addresses that gap with proprietary models for manufacturing-specific data, including two-dimensional drawings and three-dimensional CAD files. It uses general-purpose large language models for documents and spreadsheets.

The distinction defines its strategy. Instead of asking one model to understand every industrial format, the platform combines specialized interpretation with broader language capabilities.

Its manufacturing platform has four conceptual layers. The first ingests and analyzes operational data. The second adds relationships and business context. The third exposes information through search and agents. Workflow products then capture new decisions and outcomes.

That final step is important. A data platform becomes more useful when daily work continuously adds structured context. A drawing search might reveal a past defect, while a completed review records why an engineer approved the new design.

Over time, that loop can turn disconnected files into an institutional record. It can also reduce dependence on employees who remember where documents are stored or why a previous team selected a particular supplier.

CEO Yushiro Kato told Fortune that more than 80% of knowledge about manufacturing work processes is never recorded. CADDi has not provided independent research supporting that exact percentage, so it should be treated as the founder’s assessment.

The underlying challenge is broadly recognized. McKinsey found that 46% of surveyed chief operating officers cited limitations in data or IT and operational technology systems. Outdated infrastructure and poor data quality were among the specific barriers identified in its manufacturing AI research.

That finding helps explain the timing of the round. Manufacturers are experimenting with AI, but many cannot move beyond narrow pilots because their operational data remains fragmented.

CADDi is betting that the data problem should be addressed alongside the application problem. Its platform does not require customers to discard every existing system. It aims to ingest information from those systems, establish relationships, and make the combined record usable.

This approach has practical appeal. Replacing a central ERP or product lifecycle management system can become a multiyear project. An intelligence layer promises a more incremental path, starting with a workflow where slow search or duplicated effort creates visible costs.

However, layering software over fragmented systems does not automatically repair the underlying data. Duplicate part names, missing revisions, inconsistent access rules, and inaccurate records still require governance.

AI can make a connected dataset easier to query. It cannot guarantee that the dataset reflects the current factory, approved specifications, or every exception known by frontline workers.

The funding therefore arrives at a favorable moment, but also a difficult one. Enterprise buyers want usable AI now, while the information required to support it remains spread across decades of software and local practices.

CADDi Is Challenging the Fragmented Factory Stack

CADDi’s primary opponent is the fragmented factory stack, a collection of essential systems that rarely provides one usable view of engineering knowledge.

Manufacturers already own software for product design, enterprise planning, quality control, document management, procurement, and factory execution. Large vendors such as Siemens, PTC, Autodesk, Dassault Systèmes, SAP, and Oracle occupy parts of that environment.

Those vendors are not interchangeable with CADDi. Some manage authoritative product records, while others support design, planning, or production. They also continue adding AI features to their own platforms.

CADDi’s pitch is that valuable knowledge remains difficult to retrieve even after manufacturers install those systems. An engineer might search a drawing server, check an ERP record, open quality documents, and ask a senior colleague before deciding whether a previous design can be reused.

The startup wants CADDi Explorer to become the discovery layer across that process. It then wants CADDi Agent and its workflow products to take on defined analysis tasks.

For example, CADDi Design Review examines new drawings or models against problems associated with similar parts. The company says this can flag potential errors earlier and preserve lessons that would otherwise remain in individual project histories.

CADDi Agent also supports parts standardization and quality impact assessments. In those scenarios, the software must identify relevant records and help employees evaluate how a proposed change affects performance or safety.

This is a more defensible proposition than offering a generic assistant with a manufacturing interface. Technical drawings and production histories require specialized ingestion, indexing, permissions, and domain relationships.

It is also harder to deliver. The software must connect records without stripping away the revision status, product context, or access controls that determine whether information can be trusted.

CADDi’s advantage may come from deploying several applications on the same data foundation. A customer that begins with drawing search can add design review, cost analysis, or quotation workflows without building a separate dataset for each tool.

The risk is that established software providers can follow a similar path. A product lifecycle vendor already controls important engineering records. An ERP provider owns purchasing and planning context. Cloud platforms can supply storage, models, and agent-building tools.

CADDi must therefore demonstrate value between those systems. It needs to connect records more effectively than a manufacturer’s internal integration team, while delivering applications that incumbent vendors do not already provide.

Its expansion from one discovery product into six workflow products shows how it plans to defend that position. Every workflow can generate more structured data and make the central platform harder to replace.

That strategy resembles a compounding knowledge system. Search makes historical data available, applications use it during active work, and completed work creates new context for later decisions.

It also demands careful product discipline. Supporting design, procurement, quality, and manufacturing engineering can produce a broad suite before each application becomes essential.

Investors are placing a large bet on CADDi’s ability to manage that tension. The company must become broad enough to serve as shared infrastructure without becoming too diffuse to compete with specialized tools.

From Drawing Search to AI Agents

The mechanism behind CADDi’s valuation is a shift from finding files to executing decisions with manufacturing context.

Traditional enterprise search answers a location question: where is the relevant document? CADDi wants its platform to answer a relationship question: how does this design connect to previous parts, suppliers, costs, defects, and decisions?

That requires more than optical character recognition. The system must interpret text and shapes, associate related records, and preserve metadata from source systems.

A visual match between drawings can reveal that two teams ordered nearly identical components under different numbers. Purchasing data can show what each component cost and which supplier produced it. Quality records can reveal whether one version created recurring defects.

When those relationships are visible, the platform can support a business decision rather than merely retrieve a file. An engineer can evaluate reuse, a buyer can compare sourcing history, and a quality team can inspect earlier corrective actions.

The AI agent layer extends that model. An agent is software that uses models and tools to complete a defined sequence of work. In CADDi’s case, the relevant work includes analyzing standardization opportunities or assessing the effects of engineering changes.

The company’s decision to build proprietary models for drawings and CAD data addresses a real limitation. Kato told Fortune that general-purpose language models do not understand technical drawings well enough to perform design reviews on their own.

That does not mean CADDi’s models can independently approve safety-critical engineering changes. The company presents its products as tools for analysis and decision support, while human teams retain responsibility for engineering judgment.

A practical example comes from Unytite, a U.S. manufacturer of structural fasteners. The company centralized more than 60,000 drawings and production documents using CADDi Drawer.

CADDi says Unytite’s research and development teams reduced engineering and production search time by more than 90%. The published customer case also describes quality employees cross-referencing drawings, alerts, and corrective actions.

Because CADDi published the case, the claimed improvement is best read as vendor-reported customer evidence. It nevertheless illustrates why search provides an effective starting point.

The initial task is easy to understand, and a customer can measure time spent locating records before and after deployment. Search can also deliver value without immediately delegating complex decisions to an agent.

The harder phase begins when customers use the connected data to alter designs, purchasing choices, or production processes. Errors then carry greater operational consequences.

A false search result wastes time. An incomplete quality impact assessment can overlook a material risk. That difference makes traceability essential.

Users need to know which records informed an AI response, whether those records represent current revisions, and where uncertainty remains. Manufacturing AI cannot rely on plausible language alone.

CADDi’s workflow products can help by limiting agents to defined contexts and tasks. Narrow workflows are easier to evaluate than an open-ended assistant that attempts to answer every manufacturing question.

The product transition will still require evidence. CADDi must show that its agents reduce engineering lead times or material costs without increasing review burdens elsewhere.

If it succeeds, the data platform becomes more than a searchable archive. It becomes a layer through which manufacturers reuse institutional knowledge and coordinate decisions across departments.

What the Valuation Does Not Prove

A $1.2 billion valuation signals investor confidence, but it does not establish that CADDi’s expanded platform has achieved durable adoption.

The company has not disclosed revenue, customer count, retention, contract size, or the share of customers using multiple products. Those metrics are crucial for evaluating whether CADDi is becoming infrastructure or remains a collection of promising deployments.

Sales reportedly more than doubled over the past year. Without a disclosed base or revenue figure, that growth cannot be compared directly with other industrial software companies.

Headcount offers another incomplete signal. CADDi has grown to about 900 employees from approximately 600 in early 2025. That expansion supports product development and international sales, but it also raises the cost of sustaining growth.

North America will be a particularly demanding market. Manufacturers differ widely in data maturity, regulatory obligations, internal terminology, and software architecture. Integrations that work for one customer may require substantial adaptation for another.

Change management is another obstacle. Kato identified it as the largest barrier to adoption, according to Fortune. Employees must trust the results, revise familiar processes, and record decisions consistently enough to improve the shared data layer.

This creates a difficult loop. CADDi needs user participation to capture tacit knowledge, but employees are less likely to participate until the platform reliably helps them.

The company must also navigate governance questions. Engineering files may contain export-controlled information, trade secrets, personal data, supplier terms, or safety-sensitive specifications.

Connecting records increases their utility, but it also expands the consequences of weak permissions. A search or agent interface must respect restrictions inherited from several source systems.

Accuracy needs to be evaluated at the workflow level. A model can appear effective in a demonstration while missing unusual tolerances, obsolete revisions, or undocumented production constraints.

Manufacturing teams will want traceable outputs rather than unsupported recommendations. They will also need procedures for correcting data and feeding those corrections back into the system.

Competition adds more uncertainty. Incumbent engineering and enterprise software companies can place AI inside products customers already trust. Cloud providers can offer tools for building custom retrieval systems and agents.

Some manufacturers will prefer those options because they preserve existing vendor relationships. Others may develop internal systems when their data or workflows are too specialized for a shared product.

CADDi’s response is vertical integration. It combines data ingestion, domain-specific interpretation, discovery, agents, and workflow applications in one platform.

That breadth creates differentiation only if the components reinforce one another. If deployments require extensive services or manual preparation, growth may become harder to standardize.

The strongest evidence will not come from another funding announcement. It will come from customers expanding across departments and renewing after the first measurable use case.

Until CADDi reports more operating metrics, its valuation should be viewed as a claim about future platform reach. It is not proof that the company has already displaced the fragmented systems it wants to connect.

Three Signals Will Test CADDi’s Manufacturing AI Strategy

The next stage will be measured through product depth, North American adoption, and evidence that customers move from search into repeatable AI workflows.

The first signal is multi-product adoption. CADDi has expanded beyond Explorer into Agent and six workflow applications. The key question is whether existing search customers add design review, cost analysis, quotation, or production applications.

That expansion would strengthen the platform thesis. It would show that one connected manufacturing dataset can support several workflows rather than a single document search product.

Limited expansion would point in the other direction. It could mean that drawing search solves a valuable problem, while the broader applications remain difficult to deploy or overlap with existing software.

The second signal is North American customer growth. CADDi has operated in the United States since 2023 and now plans to make the region central to its global expansion.

Named deployments, renewals, and customer references will matter more than office growth or hiring. North American manufacturers provide a demanding test because they operate varied legacy systems and often impose strict security requirements.

Evidence from several industries would be especially meaningful. Automotive, industrial equipment, electronics, and contract manufacturing share data problems, but their workflows and regulatory pressures differ.

The third signal is verifiable operational impact from AI-assisted decisions. Search-time reductions are useful, but CADDi’s larger valuation depends on outcomes such as shorter engineering cycles, fewer repeated designs, lower material costs, or faster quality investigations.

Those outcomes should include clear baselines and defined measurement periods. They should also distinguish software results from broader process changes introduced during deployment.

CADDi already highlights customer outcomes on its platform, including hours saved and reductions in investigation time. More independently documented cases would make the company’s performance easier to evaluate.

Readers should also watch how CADDi describes human oversight. Movement from retrieval into design or quality decisions raises the standard for explainability, permissions, and validation.

The most credible expansion will preserve links to source records and keep accountable employees inside consequential workflows. A system that hides uncertainty behind fluent answers would weaken the case for broader adoption.

CADDi’s financing shows that investors see manufacturing data as a valuable foundation for enterprise AI. The bet is understandable: factories possess decades of knowledge that remains difficult to retrieve, connect, or reuse.

The unresolved question is whether one platform can organize that knowledge across systems without reproducing the fragmentation it was built to solve.

For engineering and operations teams, the practical response is to examine one measurable workflow first. Identify where employees repeatedly search for drawings, reconstruct past decisions, or compare similar parts. Then evaluate whether the platform improves that work while preserving traceability.

Teams building their own searchable technical repository can also review how an engineering knowledge base should handle scattered documents and retrieval.

The CADDi Series D will look justified if those initial workflows expand into a trusted data layer used across departments. If adoption stops at document search, the funding will have repriced a useful product as a much larger platform.

The next few customer deployments should reveal which outcome is taking shape.