Xometry’s AI Models Bring Manufacturing Cost and Quality Checks Into Design
- DEVELOP3D
- 988 reads
Xometry CTO Vaidy Raghavan describes plans for earlier manufacturability checks, CAD integration and AI tools for quality and supplier matching.

Manufacturing feedback is moving upstream
Xometry CTO Vaidy Raghavan sees the company’s next phase taking shape around a practical shift: giving designers manufacturing feedback while they’re still choosing a design, process and material. That means linking the company’s supplier network and pricing tools more closely to design software, while improving the models that assess parts for manufacturability, cost and lead time.

Raghavan says Xometry has spent the past year and a half expanding its presence in purchasing channels and building tools for customers pursuing larger programmes and enterprise engagements. The platform receives millions of part uploads each year, giving its models a growing base of information. He says the company also works to infer the intent behind those designs, which can help it produce more useful recommendations.
A recent example is an LLM-based process recommendation model. Xometry has offered process recommendations before, but the newer model uses AI for additional data extraction. Raghavan says the company’s design for manufacturability logic has become more accurate, allowing it to flag part risks and provide price and lead time options when a user uploads a design. He says many customers choose the recommended option, reflecting the trust they place in those results.
Designcenter puts options into the design workflow
The partnership with Siemens, announced in May 2026, is intended to put manufacturing considerations earlier in the design process. Raghavan describes this as a “shift left” approach: designers want cost, lead time and manufacturability information sooner, as product development cycles get faster. Xometry aims to make its capabilities available while a part is being designed, before a design iteration has been locked in.
Within Siemens Designcenter, users are expected to be able to explore Xometry’s full range of more than 2,000 materials and finishes. They can try different options and see the associated price as they refine a design. The value, in Raghavan’s view, is that designers can compare choices during iteration instead of discovering manufacturing risks after they’ve committed to a configuration.
Xometry has previously offered plugins for several CAD packages, including Autodesk Fusion. For now, the company is concentrating on getting the Designcenter integration and its capabilities right for the software’s users. Raghavan says the work involves establishing new habits around the tool as well as making it available. Xometry plans to serve both cloud users and customers running Siemens software on-premises, sequencing development so the company can begin testing quickly.
Scaling a global supplier network
Demand is rising across the manufacturing processes Xometry offers, according to Raghavan, without one process dominating growth. That broad demand means the company needs to keep adding manufacturing partners across sectors. He also points to increased demand for specialised certificates and requirements such as ITAR, though he gives no figures for that growth.
Expanding a supplier network across borders brings differences in engineering practice. Raghavan says European manufacturing partners may expect drawings and CAD files in specific formats, while US standards differ. Xometry has to decide which parts of its platform to standardise and where to account for regional needs.
The company’s approach is to grow local supplier ecosystems gradually into a global network. Raghavan says partners first establish themselves in their own regions, then build the capabilities and working relationships needed to serve customers internationally. That progression takes time and requires close work with suppliers.

AI needs to prove itself on complex parts
Raghavan says Xometry regularly tests new frontier models from a cold start to see how well they assess manufacturability risk and price. These models have provided more relevant information over the past year and a half, but their performance can fall sharply when part complexity increases even slightly. He also warns that hallucinations remain a serious concern, so Xometry is watching the field while waiting for greater maturity.
Some early users are exploring image-based supplier matching: provide a part image, then search manufacturer catalogues for a comparable product or supplier. Xometry is also applying language models and visual language models to work where it says accuracy is already high enough to be useful. One example combines language models with optical character recognition to detect handwriting and help generate inspection reports automatically.
Another area under investigation is live image analysis for quality management. Some manufacturing partners divide production into stages and capture images between them for immediate checks, removing faulty parts from the process before they continue through the line. Xometry’s broad supplier base gives it exposure to early adopters as well as manufacturers at different stages of adoption. Raghavan says this helps the company identify which capabilities suit which partners.
From hardware-first to data-driven production
Raghavan expects AI to push manufacturing toward a software-first model. For decades, he says, factories relied heavily on physical equipment, manual work and spreadsheets. The emerging advantage comes from processing data quickly and using machine learning to support better decisions in real time.
His longer-term view is of an AI-enabled marketplace that learns continuously from real production and feeds information back into procurement and manufacturing decisions. The practical measure will be whether these systems can produce dependable guidance across varied parts, suppliers and regional requirements. Earlier feedback in CAD, useful quality records and timely detection of defects are the areas where Xometry’s current work points to that shift.