Part 2 of Pen & Paper to Processors: nvp capital’s series on Vertical AI

One of the best parts of the East Coast is how interconnected everything is — quick flights and trains make it easy to move between major cities. After a few travel delays on the way to a board meeting in Boston, I’ve been thinking about how complex systems break down — and how often it starts upstream. In manufacturing and supply chains, something as small as a part change or supplier misalignment can cascade into real-world disruptions. It’s a powerful reminder of the opportunity for technology to bring more visibility, resilience, and adaptability to these foundational systems.

We’ve backed companies across the industrial stack — from Vulcan Elements, which is using cutting edge technology to manufacture rare earth magnets in the U.S., to Optimal Dynamics using AI to more effectively route trucks to an unannounced AI-native platform reshaping how blue-collar manufacturers are hired. We’ve spoken with execs across the industrial space and with founders building in everything from food manufacturing to fashion; and we’ve learned a tremendous amount from our LPs — many of whom operate at the bleeding edge of modern manufacturing.

At nvp capital, we’re obsessed with how software can transform the physical world — and there’s no better proving ground than manufacturing.

Why Now

Global Disruption Meets Local Fragility

Geopolitical shocks — from tariffs and COVID to U.S.–China decoupling — have exposed how brittle global supply chains really are. Teams are scrambling to diversify suppliers, onshore production, and react to ever-changing cost and compliance dynamics. Tariffs don’t just increase prices — they introduce constant uncertainty. 

Advances in Multimodal AI 

We’re particularly energized by the rapid progress in vision-language models (VLMs) — systems that can interpret visual environments, understand context, and reduce the need for manual programming and tagging. These models are rapidly improving in their ability to process unstructured, multimodal data — including raw camera feeds, thermal imagery, sensor outputs, and 3D spatial understanding — and are increasingly adaptable to messy, real-world factory settings. We’re also tracking the rise of vision-language action models (VLAMs), which convert visual context into robotic actions. Together, these advances pave the way for more adaptive, low-code automation — even in unstructured real-world environments.

Labor Shortages Are Forcing Change

Many U.S. factories are struggling to find skilled operators. One facility told us their average welder is 45 — and nationally, over 2.1M manufacturing jobs could go unfilled by 2030. Even with competitive wages, roles that require specialty skills like CNC operators and industrial welders are increasingly hard to hire for.

The result? Production timelines slip, quality control gets harder to maintain, and teams are forced to stretch limited talent across more machines and workflows than ever before.

Problem Areas We’re Excited About

System Fragmentation and Workflow Brittleness

Manufacturing is full of disconnected digital systems. A simple change — like swapping a screw or updating a CAD file — rarely syncs across sourcing, inventory, and finance without manual coordination. ERP, MES, and financial systems do not communicate, so updates get lost in email threads and spreadsheets, and downstream delays pile up. The result: decisions made today often surface as delays on the production line and financial impact weeks later, with no clear traceability.

We’re excited about teams building true interoperability — not just syncing data, but creating a connected design-to-execution loop. These systems propagate changes across tools and teams, blending automation with structured human oversight. The goal: a system that adapts in real time, no matter where or when the change originates, bringing together the entire manufacturing value chain and highlighting financial impact. 

Software That Makes Automation Actually Work

Robots and cobots are increasingly common — but deploying them in real-world settings still requires technical users. Most operators don’t have the time or training to write code or script tasks, especially in high-mix, variable environments.

We believe the real opportunity lies in software that meets operators where they are — sometimes automating repetitive tasks outright, but often augmenting human capabilities and enabling people and machines to work side by side. That might mean using natural language to train robots, real-time vision to adapt to changing conditions, or interfaces that respond to motion, voice, and intent. As Henry Rusnow, an exec at Mac Products, eloquently put it: “If someone can move something, position it, talk to it like a human — you’re off to the races.”

Understanding What’s Actually Happening on the Floor

Even as more automation hits the floor, the factory itself remains surprisingly opaque. Today, small actions — like confirming the right screw was used — often require layers of manual input, disrupting a worker’s flow. The result: people spend more time logging than doing, and most real-world activity never gets captured in a structured way.

“Good documentation of what’s happening on the floor is gold,” the heads of ops at a large-scale manufacturer told us. “How do you actually understand what they’re doing and turn it into structured data, so you can improve manufacturing efficiently?”

We’re excited about tools that capture physical activity more naturally — via scans, sensors, or video — and structure it automatically for planning, scheduling, and reporting. Operators should be able to flag issues and trigger fixes without breaking stride. Once the floor is visible, systems can adapt in real time: setups happen off the line, parts are pre-staged, machines calibrate themselves, and oversight shifts from logging to high-leverage problem solving.

Fixing the Design-to-Supplier Orchestration Layer

An often-overlooked friction point in manufacturing happens before anything hits the floor — in the messy handoff between product design and supplier onboarding. CAD files are passed around without context. Brand <> supplier communications happen across a jumble of platforms. Timelines slip, parts arrive late or samples are incorrect, and no one knows where the ball was dropped.

This challenge affects anyone building physical products — from home goods to automotive systems. There’s rarely a true system of record. Instead, teams rely on institutional memory, scattered files, and crossed fingers.

We’re excited about tools that bring structure to this chaos — solutions that:

• Centralize supplier coordination without requiring suppliers to change how they work — integrate across tools, channels, and formats behind the scenes
• Identify gaps, suggest next steps, and drive resolution across design-to-supplier handoffs — without constant human intervention
• Match buyers with the right suppliers, including visibility into upstream vendors (like tier-2 and tier-3 suppliers) to improve traceability and compliance. 

What It Takes to Win

Across all of these layers — from factory automation to supplier orchestration — some key attributes of companies that can set themselves apart. 

Fast ROI That Builds Trust

Solutions need to prove value quickly — ideally through a pilot that’s easy to implement. Teams are burned out on multi-year ERP rollouts and brittle point-solution integrations. They don’t have time to wait years for value, and operators need to see the impact on the floor to be adopted. The best products adapt to the unique requirements and constraints of each company — not the other way around.

Interfaces That Empower, Not Burden

Great UI is non-negotiable. Products must work for the people doing the work — whether that’s a welder, line operator, or factory manager. The best products make someone’s job easier, not more complicated. We love products that turn the user into the superhero — reducing downtime, increasing output, and driving revenue.

Integrate and access data: then automate

Legacy systems aren’t going anywhere. Winning products plug into what already exists, extract insights from messy data, and work across fragmented workflows. Trust is earned by integrating with finicky systems and delivering quick wins. By solving an acute, hair-on-fire problem first, companies can build credibility and expand from there — ultimately becoming the system of record or taking on the full value chain: the invisible backbone that 10x’s productivity.

In conclusion, manufacturing is entering a new era — shaped by geopolitical shocks, shifting labor dynamics, and the rapid improvement of AI. As factories face growing pressure to move faster, adapt more often, and do more with less, there’s a clear opportunity for software to bring visibility, structure, and speed to the physical world. At nvp capital, we’re investing in the next generation of vertical AI companies — ones that start with real pain points on the floor, earn trust through fast ROI, and scale into the connective tissue of modern manufacturing.

Huge thanks to Tess Hatch, Katerina Petraki, Megan Welch, Henry Russnow, and Peter Koch for their insights and guidance throughout the research for this piece.