Nidus raises $30M to put robotics AI inside defense factories

At Vannevar, Sealy worked with future Nidus co-founders Harrison Nguyen, Charu Dwivedi and Cane Punma, whose expertise the report describes as operations, robotics and AI, respectively. That combination gives the team relevant experience in hardware and restricted government environments; it does not by itself establish a manufacturing advantage.

The founders say they want to reduce U.S. dependence on China by making more use of existing domestic industrial capacity. They visited potential customers before choosing wire-harness work, where skilled workers spent hours cutting and preparing wires with hand tools. The report describes a team of roughly a dozen people and plans to hire and expand data collection in New York.

The sources differ on the founding date. The founder interview places formal founding in March 2026; a funding brief lists 2025. The available accounts do not reconcile those dates.

Existing robots, customer-specific models

Nidus's approach, as described in the financing report, is to collect proprietary demonstrations and fine-tune existing AI models for particular industrial tasks. The founders initially used open-source models and now use frontier models. A customized model could then run on a GPU alongside a robot already operating in a factory. The report does not identify the model suppliers or disclose deployment economics.

The demonstrations matter because flexible wire is difficult to manipulate consistently. During an office visit, the reporter observed teleoperation used to guide exposed wire ends against a rotating wheel to gather their strands. The team repeats such actions thousands of times to collect data. That demonstration concerns a preparation task; it does not establish autonomous assembly of an entire harness.

Data infrastructure also matters. At the reporter's October visit, the founders were waiting for a utility issue affecting a new fiber-optic connection to be fixed so they could upload training data faster to a secure cloud environment. That was a specific bottleneck at their building, not evidence of a general limit on customer deployments.

A research-review diagram outlines complementary sensing methods for robotic wire-harness assembly. It describes the wider field, not Nidus’s system. Credit: Arturo Carrera-Garzozi et al., Machines (2026), Figure 6, CC BY 4.0. Source: Machines.

Automation already exists; integration remains difficult

Wire processing already has alternatives. An October 2, 2026 research review of wire-harness robotics describes dedicated equipment for cutting, stripping and terminal attachment, as well as systems combining robotic handling, routing and other operations. Its literature search emphasized work from 2019 through August 2026 and included both direct harness research and transferable methods from adjacent fields.

The review, which draws extensively on automotive manufacturing, is not an assessment of Nidus or its defense customers. It nonetheless supplies an important comparison: individual capabilities are more developed than reliable, complete assembly sequences. Flexible wires change shape during handling, and an error in one operation can leave the next with an unusable starting configuration. Existing systems therefore narrow claims of novelty without resolving the integration problem Nidus is pursuing.

The authors also find that collaborative automation's economics depend on production volume, labor conditions and robotic cycle time. Product variation can demand new programming or workstation changes. For Nidus, the relevant comparison is the additional work its system can automate relative to equipment and human labor already available, including the engineering effort required for each deployment.

The financing report identifies broader robotics AI developers Physical Intelligence, Skild AI and Figure AI as potential rivals and notes that other startups already work on wire harnesses. The Nidus founders are betting that defense customers' precise requirements will be difficult for general robotics labs to address quickly; the report does not establish that those rivals cannot serve the market.

Nidus says its platform was built to meet defense security requirements and its AI training practices let each customer retain its knowledge. Its announcement does not establish an independently verified security certification. Aatish Nayak, the Kleiner Perkins partner who led the seed investment, argues in the financing report that private customer data and trusted access could make Nidus harder to replicate. He also suggests other robotics labs might prefer to partner with it. That is an investor's assessment of a potential advantage and business relationship, not an announced partnership.

What deployments still need to prove

The early commercial figures come from Nidus and use different measures. In the founder reporting, the company claimed more than $1 million in bookings with unnamed defense-industrial customers in four months. In its announcement, it claimed seven-figure revenue in its first four months. The accounts do not reconcile bookings with revenue or independently verify either figure.

The founders’ goal of making workers ten times more efficient is likewise an ambition described in the report, without a disclosed customer benchmark establishing that gain. The reporting does not provide deployment prices, installation costs, cycle-time comparisons, defect rates or payback periods.

The next evidence needed is production performance across repeat deployments: cycle times, accepted output, human interventions and total installation and operating costs against the customer's existing process. That would show whether fine-tuning can deliver a repeatable gain as wire types and factory conditions change. Expansion into kitting, pre-assembly and printed circuit boards remains an ambition described by the founders; the current reporting does not establish those capabilities.