Why I Joined Silk: The AI Inference Infrastructure Opportunity

I’ve spent my career at the intersection of data and enterprise technology — SingleStore, TIBCO, Conga — and I’ve developed something of a sixth sense for the moments when infrastructure and market demand snap into alignment. When they do, the commercial opportunity doesn’t just appear. It arrives all at once.

This is one of those moments. And Silk is sitting directly in the gap it creates.

The inference gap nobody talks about

The AI conversation has been dominated, understandably, by models. Who has the best LLM. Which lab is winning the benchmark race. But the war is increasingly being won at the infrastructure layer — and specifically at the point where AI inference meets data.

Here’s the problem: AI inference isn’t like traditional compute. When an AI model is serving real-time decisions — fraud detection, medical diagnostics, personalized recommendations, agentic workflows — it needs data immediately. Not in milliseconds. Not after a cache miss. Right now. The model is only as good as the speed and freshness of the data flowing through it.

“The gap between what AI inference demands and what general-purpose cloud infrastructure can deliver — that gap is widening, and it’s costing enterprises real money.”

The dirty secret of enterprise AI deployments is that companies are spending billions on GPU clusters and frontier models, then bottlenecking the entire investment on aging data infrastructure that was designed for a completely different era. The model waits. The business suffers. And the promised ROI of AI never fully materializes.

Why now, and why Silk

I’ve seen a lot of technology cycles. The shift from on-prem to cloud. The rise of data lakes. The streaming revolution. Each of these opened up generational commercial opportunities for the right platform at the right time. AI inference is the next one — and it’s bigger than any of them.

• 100× faster performance for mission-critical workloads
• 50% lower cost vs. general-purpose cloud infrastructure
• 0 compromises on performance when data powers everything

When I looked at Silk, I saw something rare: a platform that was purpose-built for exactly the workload that is becoming the dominant use case in enterprise computing. General-purpose cloud infrastructure wasn’t designed for the data demands of real-time AI inference. Silk was. That’s not positioning. That’s architecture.

Silk eliminates the resource constraints and latency penalties that hold other platforms back — and does so at a cost structure that makes the business case easy. In a world where CFOs are scrutinizing AI spend and demanding measurable returns, that combination of performance and economics is genuinely differentiated.

The energy problem is becoming the AI problem

There’s another constraint that doesn’t get enough airtime in the infrastructure conversation, but that every serious operator is quietly losing sleep over: power.

AI inference is extraordinarily energy-intensive. Data centers are running into hard limits — not because the technology isn’t ready, but because the grid can’t keep up. Hyperscalers are signing nuclear power agreements. Neoclouds are fighting for megawatts. Regulators in multiple jurisdictions are scrutinizing data center energy consumption. The ability to do more with less power isn’t just a sustainability story — it’s rapidly becoming an operational prerequisite for anyone building AI infrastructure at scale.

ENERGY & EFFICIENCY

Silk’s architecture dramatically reduces the storage I/O overhead that makes traditional data platforms so power-hungry. By decoupling compute from storage and eliminating redundant data movement, Silk enables AI inference workloads to run with significantly lower energy draw — delivering the same or better performance at a fraction of the power cost. For neoclouds operating at the edge of their power envelopes, that efficiency isn’t a nice-to-have. It’s a strategic unlock.

This is why the neocloud opportunity is so compelling to me. These are the operators building the next generation of AI infrastructure — purpose-built clouds optimized for inference workloads, running closer to the edge, with tighter power budgets than the hyperscalers. They are under more pressure than anyone to squeeze maximum performance out of every watt they consume. Silk’s efficiency profile maps directly onto their most pressing constraint.

When a neocloud can tell its customers that their AI inference workloads run 10x faster, cost 50% less, and consume materially less energy — that’s not a feature comparison. That’s a competitive moat. And Silk is the platform that makes it possible.

The commercial opportunity is enormous

What drew me here isn’t just the technology — it’s the market structure around it. The explosion of AI inference is creating new constituencies of buyers: neoclouds, hyperscalers, and enterprises deploying AI at scale who are discovering, sometimes painfully, that their existing data infrastructure is the limiting factor. Increasingly, their energy bills are making that lesson even more expensive.

My mandate is to build the partnerships and commercial frameworks that connect Silk’s platform to these buyers at speed. That means strategic alliances with the infrastructure players building the AI stack. It means being the data platform of choice for neoclouds that need performance and efficiency in equal measure. And it means being the platform enterprises call when their AI ROI is being held hostage by their data layer — or their power bill.

“AI inference is the defining infrastructure battleground of this decade. Silk has cracked the code on performance, cost, and now energy — at exactly the moment when all three matter.”

What I believe

I believe we are at the beginning — not the middle, not the end — of the AI infrastructure buildout. The models will keep improving. The inference workloads will keep multiplying. The data demands will keep accelerating. And the pressure on power and energy efficiency will only intensify as deployments scale. Every one of those trends is a tailwind for Silk.

I also believe that the winners in infrastructure cycles are rarely the companies with the most features. They’re the companies that understand the specific, acute pain their customers feel — and eliminate it decisively. Silk understands that pain across three dimensions that matter right now: performance, cost, and energy. No general-purpose platform can match that combination.

That’s why I joined. The opportunity is real, the timing is right, and the technology is ready. I’m excited to build the business around it.