The Bottleneck Nobody is Hiring For

Recently, I spoke with the CSO of one of Europe's most advanced photonics companies. They spent twelve years and $600 million building display technology. They are not abandoning it, but they are looking at the AI datacentre market and seeing something far larger. "The infrastructure buildout is the real opportunity," they told me, completely matter-of-factly.

Then they added a detail most missed. Jensen Huang does not get excited easily. When he stood on stage at Computex and called Marvell the next trillion-dollar company, the room nodded politely and moved on. They were too busy fantasizing about frontier LLMs and autonomous software agents. Almost nobody noticed that the CEO of Nvidia — a man who has spent thirty years thinking about what compute actually requires to function — was pointing directly at a connectivity semiconductor business.

The CSO had noticed. That gap in attention is where the real story lives.

I am not a semiconductor analyst. My job is placing C-suite and board leaders across DeepTech and AI infrastructure companies in Europe. This puts me in rooms where strategic direction is decided before it becomes public knowledge. The pattern is clear: companies with deep hardware capability are quietly repositioning around AI infrastructure because the market signal is simply too large to ignore.

It points toward a leadership crisis that almost nobody in the mainstream AI conversation is talking about yet.

Connecting hundreds of thousands of GPUs into a single coherent system is not a software issue. It is a physics problem. And physics has a brutal, uncompromising answer: light moves faster than electrons, sips less power, and does not degrade over the distances that modern AI datacentres require.

Beyond ten meters, traditional copper interconnects melt under the bandwidth demands of modern AI factories. The entire industry is now in a frantic race to replace copper with optical processing at every layer of the stack: from rack to rack, chip to chip, and eventually within the package itself.

From Nvidia's definition of the AI Factory, to Meta's 90-million-hour reliability validation for co-packaged optics, to Broadcom pushing co-packaged optics toward commercial maturity, the entire sector is solving for the same constraint. As clusters scale toward one million GPUs, the interconnect dictates the architecture. Marvell's sudden prominence is not a coincidence.

At Computex, Marvell unveiled its Teralynx T100, a 102.4 Tbps switch silicon purpose-built for AI infrastructure. It cuts power by 25% and slashes latency for intense training workloads. Huang's stage endorsement was not a courtesy. It was a warning flare about where the actual constraint sits: not in the models, but in the connectivity layer that binds them into a system.

Marvell did not build this end-to-end portfolio to be polite. They built it because every hyperscaler, sovereign compute cluster, and AI factory on earth is about to hit the same physical wall.

This is picks-and-shovels infrastructure at a planetary scale. The broader market is still staring at the chatbots. The smart money is staring at the switch silicon.

Across France, Germany, and Switzerland, companies that spent a decade building foundational hardware in photonics, semiconductors, and advanced materials are suddenly sitting in the direct path of the largest capital buildout in computing history. Some are pivoting deliberately; others are being pulled there by hyperscalers desperate to diversify supply chains away from Asia.

Europe possesses world-class depth in this layer of the stack. Historically, our talent lay in inventing magnificent things in quiet laboratories, then watching American companies commercialize them at speed.

That historical patience is now a liability. The technical advantage is here, but the commercial velocity is missing.

The datacentre sector is already facing a severe shortage of specialized professionals. But while the industry frets over engineers and technicians, the true execution bottleneck sits directly in the C-suite.

You cannot retrain a traditional SaaS executive to negotiate multi-year product design cycles with Nvidia or manage capital-intensive semiconductor manufacturing runs. The profiles required are fundamentally different. They demand deep hardware intuition, tolerance for long development cycles, and the ability to sell into the monolithic procurement teams of the world's largest technology companies.

The CTO who can scale an optical interconnect company from an academic lab to a multi-million-dollar hyperscaler contract does not browse job boards. The CPO who understands silicon photonics well enough to map a roadmap, but remains sharp enough to close a design win at Microsoft, does not have a public resume.

They are already running a company, advising a sovereign fund, or being quietly shielded from the market by investors who identified them three years ago. Finding them takes actual investigative work, not an automated LinkedIn sequence.

The $700 billion buildout is underway. The leadership race to support it has barely started.