The story of EV adoption contains a warning that semiconductor companies would be wise to take seriously. When the automotive industry made the leap from combustion engines to electric drivetrains, it discovered that a transformative technology is not enough on its own. The ecosystem supporting it has to be ready too. Today, the semiconductor sector is standing at an almost identical inflection point.
Advanced silicon is moving fast. The software infrastructure required to deploy it at scale is not.
A Pattern the EV Industry Knows Well
For over a century, automotive progress was built on hardware — engines, drivetrains, safety systems — with software serving a supporting role and broader systemic questions around fuel economics and refueling convenience treated as secondary concerns. Electric vehicles rewrote the propulsion model entirely, but early manufacturers largely stayed with the old playbook: engineer impressive hardware, add just enough software to make it operational, and assume the wider ecosystem would catch up in time.
The consequences were predictable. Research from UC Berkeley points to potential economic and environmental gains in the trillions if EVs reach widespread adoption — yet uptake moved far more slowly than anticipated. Charging infrastructure was unreliable, wait times were long, and the total cost of ownership failed to convince the average driver. The technology had changed, but the complete, software-driven experience that would actually shift behavior had not been built. Most consumers stayed anchored to what had worked for a century.
Semiconductor Companies Are Repeating the Same Mistake
The parallel with AI inference is direct. Semiconductor companies are engineering powerful, purpose-built accelerators, NPUs, and AI-focused SoCs — the equivalent of next-generation electric motors. But without deep, vertically integrated software stacks covering compilers, runtimes, orchestration, observability, and edge-aware deployment tooling, the AI equivalent of reliable charging infrastructure simply does not exist.
There is a second dimension to this gap. Semiconductor companies need to work closely with their customers to identify the specific high-value use cases — typically a handful — that will actually drive meaningful demand. These applications are what determine whether a technology reshapes the value chain or remains on the periphery. Without them, demand builds slowly and significant upfront investment generates limited real-world return.
Increasingly, it is software that determines which hardware platforms win.
Procurement decisions are shaped less by the chip itself than by the quality of the ecosystem surrounding it — making the two effectively inseparable. The dynamic mirrors range anxiety in EVs: drivers do not evaluate the vehicle in isolation, they think about where they can charge it and whether that works for their life.
Where the Value Actually Gets Created
Just as EV adoption stalled until drivers could charge reliably at home, at work, and on the road, enterprise AI adoption will remain constrained until organizations can deploy, monitor, and cost-effectively run models at the edge — close to where their data is actually generated. When software stacks are thin or fragmented, inference costs stay high, operational complexity compounds, and AI remains a compelling demonstration rather than an embedded capability within daily products and workflows.
Decades of semiconductor investment risk underdelivering for the same reason early EVs did: the end-to-end experience and cost structure have not been optimized for scaled, everyday use.
Companies shipping advanced AI silicon without mature software ecosystems are doing what early EV manufacturers did — releasing impressive technology into an environment that is not yet equipped to support it at scale.
The companies that treat software as the charging infrastructure for their silicon — making edge inference straightforward, economical, and transparent — are the ones positioned to see faster adoption, deeper integration into customer products, and a larger share of where long-term AI value ultimately lands.
Here are a few versions:
Craig Melrose is Global Managing Partner, Advanced Technologies at HTEC. With over 30 years of experience spanning Toyota, McKinsey, and PTC, Craig specializes in translating operational complexity into measurable business outcomes across the semiconductor and advanced technologies sectors. To continue the conversation, reach out to Craig directly.
