Vesence Thinks AI Can Finally Fix Chemical Manufacturing's Oldest Problem

The Macro: Factories Are Running on Instinct

Chemical manufacturing is a $5.7 trillion global industry, and a surprising amount of it still runs the way it did twenty years ago. Plant operators monitor batch processes, adjust parameters based on experience, and make judgment calls that can mean the difference between a clean run and thousands of dollars in wasted feedstock. The knowledge lives in people’s heads. When veteran operators retire, it walks out the door with them.

This isn’t a secret. Everyone in process manufacturing knows about the tribal knowledge problem. The reason it persists is that the software alternatives have been either too rigid (traditional process control systems that automate fixed sequences but can’t adapt) or too disconnected from the actual chemistry happening on the floor.

The industrial AI market is growing fast. Multiple analyst estimates put it north of $70 billion by 2030, with process optimization being one of the most obvious applications. But the gap between “AI could help here” and “AI is actually deployed on this production line” remains enormous. Most factories are still in the pilot phase, running limited experiments with vendors who overpromise and under-deliver on integration.

The competition in this space includes established players like AspenTech, which has been doing process simulation for decades and has deep relationships with chemical companies. Siemens has industrial AI ambitions. Honeywell has been making similar noises. There are also newer entrants like Sight Machine and Uptake, focused on different slices of manufacturing intelligence. The incumbents have distribution but slow product cycles. The startups have speed but lack domain credibility.

Vesence is trying to split the difference: a startup with genuine manufacturing domain expertise, building an AI operating system purpose-built for process plants.

The Micro: From Basketball Courts and Recording Studios to Factory Floors

The founding team here is genuinely unusual. Henrik Hansson, the CEO, studied physics and economics at two of Sweden’s top universities simultaneously. He was also a music producer with ongoing royalties, which is the kind of biographical detail that makes you do a double take when you’re reading about chemical plant software. Ludvig Swanstrom, his co-founder, was a professional basketball player before transitioning into tech. Both of them spent three years at Depict, a YC S20 company, working alongside the founder who later built Lovable. They came through Y Combinator’s Spring 2025 batch.

The product pitch is straightforward: Vesence integrates with existing plant systems (historians, SCADA, DCS) and builds AI models that understand the specific chemistry and physics of each production process. Instead of generic optimization recommendations, the system learns the relationships between raw material variability, equipment conditions, and output quality for a specific plant’s specific products.

The core value proposition breaks down into two pieces. First, throughput optimization. If you can predict how process variables interact in real time, you can push production rates higher without sacrificing quality or safety. Even small percentage improvements in throughput translate to millions of dollars annually at scale. Second, waste reduction. Off-spec batches, rework, and material losses are endemic in process manufacturing. If AI can catch deviations early and recommend corrections before a batch goes bad, the savings compound quickly.

What I find interesting about the positioning is the “operating system” framing. They’re not pitching a point solution for a single optimization problem. They’re betting that once you have a real-time AI model of a plant’s processes, you can layer multiple applications on top of it: quality prediction, energy optimization, yield improvement, maintenance scheduling. That’s an ambitious scope for an early-stage company, but it’s also the kind of platform bet that creates defensibility if they can execute.

The practical question is implementation complexity. Every chemical plant is different. Different equipment, different chemistries, different legacy systems, different regulatory environments. Pharma plants have FDA validation requirements that make software changes painful. Food processing plants have their own set of safety and compliance constraints. Building something that works across chemicals, pharma, and food manufacturing is three separate integration challenges wearing a trench coat.

The Verdict

I think the problem is real and the market is large enough to support multiple winners. The tribal knowledge gap in process manufacturing is getting worse, not better, as experienced operators age out of the workforce. AI-powered optimization is the obvious answer, and the question is really about execution.

At 30 days, I’d want to see a live deployment at a single plant with measurable throughput or yield improvements. Not a pilot. Not a proof of concept. An actual production system making real-time recommendations that operators follow.

At 60 days, the integration story matters. How painful is it to connect to existing plant systems? How much custom modeling is required per site? If every deployment is essentially a consulting engagement, the economics get difficult fast.

At 90 days, the question is whether the “operating system” ambition holds up or whether Vesence needs to narrow its focus to one industry vertical and one use case to build enough depth to win. AspenTech has had decades to build its position. A startup needs to find one wedge that’s undeniably better and drive it deep.

The team’s background is unconventional, but their time at Depict gives them startup operating experience, and the YC network provides distribution advantages in fundraising and early customer intros. Whether a physics-major-slash-music-producer and a former pro basketball player can win the trust of chemical plant managers is its own kind of challenge. But I’ve seen stranger founding stories lead to serious companies.

Worth watching. The space needs fresh thinking, and the incumbents are slow.