7 Go-to-Market Lessons from Building AI Chips: What SEMRON Learned Swimming in the Ocean for Four Years

Your VP of Sales wants to know the pipeline forecast. Your investors want to see month-over-month growth. And you’re staring at a product roadmap measured in years, not sprints. In a recent episode of Category Visionaries, Aron Kirschen, CEO of SEMRON, an AI chip maker that’s raised $7.3 million, shared what most hardware founders learn the hard way: everything you know about go-to-market doesn’t apply when physics determines your timeline.

Lesson 1: Redefine What “Scale” Means for Your Business

Silicon Valley worships scale. More users, more revenue, more growth. But Aron operates under a completely different definition of success. “We do not or will not likely won’t have millions of customers. For us, it’s more like having five, six big entities that is generating millions of revenue.”

Five customers generating millions in revenue inverts the traditional scaling playbook. There’s no growth marketing team optimizing funnels. No demand generation engine churning out MQLs. Instead, it’s about identifying the right five companies and becoming essential to them.

This isn’t a limitation. It’s strategic clarity. When you know your total addressable customer count maxes out at single digits, you stop worrying about scalable acquisition channels and start thinking about relationship depth. You allocate resources differently. You hire differently. You measure success differently.

For hardware founders, the lesson is simple: figure out your actual customer count, not your aspirational one. If you’re selling to five enterprises, build a GTM motion for five enterprises. Don’t pretend you’re building the next PLG SaaS company.

Lesson 2: Match Your Timeline to Your Technology

Most startup advice assumes a 12-month product development cycle. Ship fast, iterate, repeat. But Aron lives in a different universe. “As a semiconductor startup, you don’t have the timeline or the roadmap usually of 1 year for go to market to really have to think in three, four, even five years.”

The timeline depends entirely on what you’re building. Working with existing semiconductor technology and just designing a chip? One to two years. Starting with a new semiconductor device like SEMRON? Three to five years because “you really have to open the semiconductor process and make some changes there together with the foundry.” New materials beyond standard silicon? Eight to ten years.

This creates a fundamental mismatch with typical investor expectations, hiring practices, and team morale strategies. You can’t promise customer traction in six months when you’re two years from having a demonstrable product.

The solution isn’t to lie about timelines. It’s to build a company culture and investor base that understands deep tech timelines. SEMRON is four years in and “close to process freeze”—the moment when they can finally start real go-to-market work. That’s not slow. That’s physics.

Lesson 3: Your Product Is the Demonstrator

In SaaS, you have free trials and freemium tiers. In consumer apps, you have viral loops. In AI hardware, you have one thing that matters: demonstrators. “For us, it all comes down to these demonstrators going out to customers, showing them our FPGA together with our hardware,” Aron explains.

But building demonstrators in 2024 means solving a problem that semiconductor companies didn’t face a generation ago: “Even though we are hardware company, so today hardware means you have to do a lot of software.”

SEMRON’s process reveals what sophisticated B2B sales actually looks like. They receive proprietary AI models from potential customers—the actual workloads these companies need to run. Then they deploy these models on their hardware, which they currently emulate, and “come back with a nice performance simulation.”

This isn’t marketing collateral. This is engineering as sales. The demonstrator proves value in the most concrete way possible: running the customer’s actual models on your actual hardware and showing actual performance data. No hand-waving allowed.

Lesson 4: Embrace the Ocean Years

Perhaps the most honest moment in Aron’s story comes when he describes the early years: “For a semiconductor startup that for three, four years worked like it feels, it felt like swimming in the ocean. And you never had the real feedback.”

Those ocean years—when you’re building without customer validation—are unavoidable in hardware. The market feedback you get is superficial. Potential customers saying “we might be interested, blah, blah.” Investors doing diligence calls to gauge whether your technology has potential. None of this tells you what you actually need to know.

The breakthrough comes when you finally reach real customer engagement. “It’s really like we understand what they need in an application. And of course it completely reshapes your coordinate system. You never thought about that before,” Aron admits.

For hardware founders, this means two things. First, accept that the ocean years exist. Don’t fight them or pretend they’re not happening. Build your burn rate, team structure, and fundraising strategy around swimming in the ocean for years. Second, when you finally reach shore—when real customer conversations start—be ready to have everything you thought you knew completely reshaped.

Lesson 5: Engineering Discussions Are Your Sales Motion

Forget the sales playbook. In SEMRON’s world, sales looks completely different. “It’s very intense, very technical, very detailed, of course, but it is less standard marketing. It’s not that much based on emotions, convincing people to use it. I would say it could be different, but in our case it’s really easy.”

The entire sales motion comes down to one action: “You have to sit down with their engineering team, with our engineers, and figure it out how we can make it work.”

This isn’t a meeting. It’s not a demo. It’s a deep technical collaboration where both sides bring their expertise to solve a specific problem. Your engineers need to understand their architecture, their constraints, their performance requirements. Their engineers need to understand your capabilities, your limitations, your roadmap.

This approach only works when you’re targeting the right customers—those five or six major entities. You can’t scale this motion to hundreds of customers. But you don’t need to. You need it to work perfectly for five companies. That’s the entire business.

Lesson 6: Cost Economics Trump Technical Performance

When Aron talks about what actually matters for their GTM strategy, he doesn’t lead with performance specs. He leads with cost. “I know that’s an important thing, but you also have to talk about cost, right? Cost per compute.”

The math is brutal and simple. Even if you could make an energy-efficient AI solution using existing technology, “we would not be able to make a product out of it because nobody would pay for like $5,000 more for something like that or even $200 more.”

The target? “We really have to be like let’s say 20, $30.”

This cost constraint fundamentally shapes the entire product strategy and, by extension, the go-to-market approach. SEMRON isn’t selling marginal improvements in performance. They’re selling economic viability for an entirely new category of devices. The value proposition isn’t “10% better performance.” It’s “this category of products can finally exist at all.”

Understanding your true economic constraint changes everything about how you position your product, who you sell to, and what conversations you have with customers.

Lesson 7: Customer Engagement Beats Customer Acquisition

As SEMRON moves into their next phase, Aron’s focus crystallizes around a single objective: “It’s about customer engagement.” Not acquisition. Not activation. Not expansion. Engagement.

For the first time after four years of development, SEMRON is having substantive conversations with “very big, well known companies that are market leaders in their application or in that vertical.” They’re receiving proprietary models, running real benchmarks, and proving out performance against actual requirements.

This shift from building in isolation to engaging with specific customer problems represents the true beginning of go-to-market for hardware companies. Everything before this—the patents, the process development, the test structures—was prerequisite work. The actual market entry starts when you’re sitting across from a customer’s engineering team, solving their specific problem with your specific technology.

For hardware founders navigating similar journeys, SEMRON’s path offers a crucial insight: your go-to-market motion doesn’t begin when you incorporate or even when you have a working prototype. It begins when you can engage meaningfully with customer problems. Everything before that is just getting ready to start.