Every medtech conference this year has an AI keynote, and most of them are about the same thing: smarter diagnostics, faster imaging, algorithmic triage. That's real, and it matters. But it's not where I think the bigger shift is happening for medical device and capital equipment companies specifically. It's also, increasingly, where we at Chakra think some of the more interesting investment opportunities sit.
Before venture investing, I spent over a decade in diagnostics and medical device businesses, running global product portfolios and, before that, scaling a service and software business at a capital equipment company. Now, evaluating AI and software companies as an investor, I look at this same shift from the other side of the table. Both views point to the same conclusion: incumbents have something a new AI-native entrant cannot easily replicate — a large, connected installed base generating years of real usage, failure, and service data. That data is the actual raw material for AI in this industry, and most companies are sitting on far more of it than they're using.
Customer needs are hiding in service data, not just surveys
The traditional way to find unmet customer needs is a voice-of-customer program: interviews, KOL panels, journey mapping. Those are still essential — I've built and run these programs, and there's no substitute for talking to the people using your equipment. But they're slow, and they only surface what customers can articulate.
AI applied to machine analytics changes that. When you can pull performance and usage trends off an installed base — as the more connected device companies are now able to do — you start seeing patterns customers themselves haven't named yet: which failure modes cluster before a formal complaint ever gets filed, which usage patterns predict a service call three weeks out, which regions are quietly struggling with a workflow the sales team assumed was fine. That's the unmet need showing up in the data before it shows up in a survey. For companies managing thousands of complex instruments across dozens of countries, that's not a nice-to-have — it's the only way to actually keep up with the fleet.
Field service is becoming a software product, whether or not it's labeled that way
The other shift is that "asset management" for medical equipment — monitoring, maintaining, repairing a customer's installed base — is turning into a software and data problem as much as a wrench-and-parts problem. Predictive maintenance, remote diagnostics, digital service scheduling, parts logistics driven by failure forecasts: these all require the same muscle that a software product team uses — defining requirements, running structured launches, working through UAT, iterating with real usage data. Companies that already run rigorous product lifecycle processes for their hardware have most of the organizational muscle to do this well; they just haven't always pointed it at the service side of the business with the same discipline.
This is also where the human factors and regulatory realities of medtech actually become an advantage rather than friction. Every AI-driven service or diagnostic tool touching patient-adjacent equipment has to hold up under FDA scrutiny, clinical workflow constraints, and genuine safety stakes. That's a real barrier to entry. A startup can ship a slick predictive model; shipping one that a hospital's biomed team, a regulatory body, and a clinician can all trust is a different exercise entirely, and it's one that established device companies have already been doing for years with things like 510(k) submissions and post-market surveillance. The advantage isn't just data — it's the operating discipline to turn that data into something safe, validated, and scalable.
Why this is a capital allocation question, not just an operating one
At Chakra, this shift shapes how we think about diligence, not just how operators should think about their roadmaps. When we evaluate an early-stage company applying AI to a traditional industry, the questions we ask are the same ones an operator should be asking internally: Is there a real, defensible data asset here, or just a model wrapped around public data anyone can access? Does the team understand the regulatory and safety constraints well enough to actually ship, or are they underestimating how long validation takes in a clinical setting? Is the customer problem real and costly enough — measured in downtime, service cost, or missed revenue — to justify the investment, or is this a solution in search of a need?
Those questions matter whether you're a corporate venture arm deciding where to place a strategic bet, an incumbent deciding whether to build or partner, or a founder trying to figure out whether their wedge is defensible. The installed-base advantage that makes incumbents hard to disrupt is exactly the kind of asset that makes a partnership, licensing deal, or minority investment more interesting than a head-on competitive play — for both sides of that table.
Where the opportunity actually sits
None of this argues for chasing AI as a headline feature, or investing in it as one either. The companies — and the incumbents — that will win this transition are the ones that treat it as a continuous-improvement problem: define success metrics up front, build the business case with real qualitative and quantitative evidence, pilot in a contained way, and only scale what measurably improves uptime, cost, or the customer experience. That's Lean thinking applied to a new toolset, not a moonshot, and it's also the kind of discipline we look for when evaluating a company from the outside.
The organizations best positioned to do this — as operators or as investment targets — are the ones with three things already in place: a large connected installed base, cross-functional teams that can move between hardware, software, and field service without territorial friction, and a genuine habit of listening to customers and translating that into the roadmap rather than treating service as a cost center to be minimized. Most incumbents already have the first ingredient by default. The differentiator, and the thing worth underwriting, is whether they build the second and third on purpose.
That's the opportunity we find most interesting right now, from either seat — not AI as a shiny new feature or a hot thesis, but AI as the thing that finally lets service, product, and customer insight operate as one connected system instead of three separate functions comparing notes after the fact.