A record that arrives quietly, but speaks volumes to those scrutinizing the foundations of artificial intelligence. WITS, a company specialized in custom system-on-chip design, has announced its highest-ever revenue for a first half. It’s not a mere insider detail: it’s a litmus test of a broader shift that moves the center of gravity for inference from centralized data centers to solutions closer to the data.

The stated drivers — artificial intelligence and chip design — are not separate tracks but an intertwined whole. Demand for custom silicon for inference workloads is growing because enterprises are pushing processing to where latency is minimal and data control is absolute. It’s not just a performance issue: it’s an architectural choice that impacts Total Cost of Ownership (TCO) and regulatory compliance, especially in Europe.

The business model of companies like WITS rests on providing design expertise to clients that lack their own fabs or internal microelectronics teams. This lowers the barrier for those wanting to develop ASICs dedicated to LLM inference, perhaps optimized for a specific model family or for aggressive quantization. The result is an ecosystem where even a manufacturer or a healthcare operator can evaluate silicon tailored to their needs without incurring the cost of building a design team from scratch.

The structural signal is twofold. On one hand, it confirms that the market is not flattening around GPUs as the only solution. AI hardware acceleration is diversifying, and the growth of design service providers bears witness to that. On the other, it reinforces the notion that on-premise and edge deployments are becoming economically and technically viable even for mid-sized enterprises, not just for big tech. Data sovereignty is no longer a slogan: it’s a requirement that translates into custom chip projects, often designed to run in air-gapped environments or under strict data residency rules.

For those surveying the landscape with an eye toward local stacks, WITS’s figures are a leading indicator. Behind the scenes, the semiconductor supply chain is already responding to demand for hardware that is not only powerful but also suited to regulated contexts, with attention to energy consumption and real operating costs. And while the big names in AI monopolize the spotlight, the real game is being played in the design labs where the silicon that will run tomorrow’s models is being shaped.