DON’T HIT GRANDMA IN THE CROSSWALK

🚨 A Waymo nearly hit me in a San Francisco crosswalk.My first thought was not about litigation.It was: this vehicle needs faster, better inference.Luckily, I was quick enough to get out of the way. Grandma may not be.Training teaches an autonomous vehicle what a pedestrian looks like. Inference determines whether it stops before hitting one.That distinction sits at the center of our investment thesis for a $10 billion AI infrastructure strategy.The first phase of the AI boom was dominated by massive training clusters and multi gigawatt campuses. Those facilities remain essential. But the next phase will be driven by what happens after the model is trained.Every fraud alert, medical interpretation, robotic movement, industrial control, customer interaction, and autonomous vehicle decision requires inference.And inference has a different physical geography.⚡ It needs power that can be secured quickly.📍 It often needs to operate closer to customers, machines, and regulated data.🏗️ It benefits from modular facilities that can be built in financeable phases.🌐 It favors distributed neocloud infrastructure rather than total dependence on a handful of hyperscalers.🤝 It can reduce the political and community concentration risk created by enormous speculative campuses.The market is beginning to validate this architecture. EPRI, NVIDIA, Prologis, utilities, and infrastructure developers are now evaluating distributed inference centers in the 5 to 20 megawatt range near substations with available power.That is not edge computing theater. It is institutional infrastructure.The problem is that these projects remain difficult to finance. They are too physical for traditional venture capital, too operational for conventional real estate lenders, too early for core infrastructure funds, and often too specialized for generalist capital.That financing gap is not a weakness in the thesis.It is the opportunity.Our view is straightforward: the future is not large data centers versus small data centers. The future is a layered compute network.Large campuses will train frontier models. Regional and modular neocloud centers will deliver inference closer to the economic activity those models serve.The winning portfolio will not simply own the most GPUs.It will control the right power, in the right locations, supporting the right workloads, under contracts that convert artificial intelligence into durable cash flow.Because I may still be fast enough to outrun a confused autonomous vehicle.Grandma should not have to be.Read “Don’t Hit Grandma in the Crosswalk: Why the Future of AI Compute Is Inference and Smaller, More Nimble Neocloud Data Centers.”#ArtificialIntelligence #DataCenters #Inference #Neocloud #AIInfrastructure #PowerInfrastructure #SovereignAI #PrivateCapital #InvestmentBanking #AutonomousVehicles #FutureOfCompute #GROGroupImportant NoticeThis paper is provided solely for informational and discussion purposes. It does not constitute an offer to sell or a solicitation of an offer to purchase any security or investment product. Any investment opportunity will be offered only through definitive offering documents and in accordance with applicable securities laws. Market forecasts and projections are inherently uncertain and should not be interpreted as guarantees of future results. Prospective investors must conduct their own legal, financial, tax, commercial, and technical diligence.(c) 2026 Grobecker Holland International, INC. All rights reserved.