# Neil Pinto on enterprise AI's next phase > LinkedIn's former infrastructure chief explains why compute economics, not technology, will unlock enterprise AI adoption. - Title: Neil Pinto on enterprise AI's next phase · Futuriom | Shortify - Summary: LinkedIn's former infrastructure chief explains why compute economics, not technology, will unlock enterprise AI adoption. Enterprise AI adoption will… - Keywords: neil pinto, linkedin, enterprise ai, gpu compute, ai infrastructure, inference costs, ai adoption, compute economics, ai roi, business outcomes, neoclouds, hyperscalers - Source: Futuriom — https://futuriom.com/articles/news/q-a-neil-pinto-vp-infrastructure-engineering/2026/09 - Read time: 4 min - Topics: neil pinto, linkedin, enterprise ai, gpu compute, ai infrastructure, inference costs, ai adoption, compute economics, ai roi, business outcomes, neoclouds, hyperscalers ## From experimentation to production Enterprises have moved past asking if they should use AI to asking where it creates measurable value. Hundreds of pilots across customer service, software development, and other areas show clear productivity gains. ## The scaling problem is expensive GPU compute remains constrained and inference costs are significant for large deployments. Enterprises are uncertain which models and strategies make economic sense before committing to production. ## Lower costs unlock adoption As GPU capacity expands and cost per token declines, enterprises will shift from optimizing for experimentation to optimizing for business outcomes. That's when adoption will accelerate significantly. ## Value moves up the stack Today most spending is on infrastructure like GPUs and datacenters. Over time, value will migrate to AI applications, autonomous agents, and industry-specific solutions, similar to how cloud computing evolved. ## Still early innings for AI More available compute, lower inference costs, and greater confidence in AI ROI will unlock the next wave. Demand will broaden from frontier model companies to thousands of enterprises deploying AI. ## Economics is the first barrier Every AI interaction has a cost. Enterprises constantly evaluate cost per token against business value created. As GPU capacity increases and inference costs decline, many more use cases become economically viable. ## Skills and execution matter most Deploying AI requires data engineers, AI infrastructure experts, and people who understand governance and workflow orchestration. Companies that succeed will integrate AI into daily operations, not just buy GPUs. ## Workforce fears are real Employees worry AI will make their roles obsolete. History shows technology reshapes jobs more than eliminating them. Companies that position AI as augmenting employees, not replacing them, will succeed. ## Models aren't the bottleneck Incredibly capable models already exist. The real challenge is turning them into end-to-end enterprise solutions that deliver measurable business value across connected workflows. ## Point solutions create integration chaos Startups are solving specific problems like customer support or coding. But enterprises now face integrating dozens of AI tools into a coherent operating model, creating opportunity for systems integrators. ## Don't measure just infrastructure High GPU utilization doesn't mean high business value. The real ROI comes from what tokens accomplish: faster customer service, more productive engineers, reduced costs, and improved outcomes. > High GPU utilization doesn't necessarily mean high business value. ## Track two layers of metrics Infrastructure KPIs include GPU utilization and cost per token. Business KPIs measure productivity gains, revenue, cost reduction, and customer satisfaction. Winners connect both layers together. ## Tokenomics isn't the real metric Cost per token is like measuring electricity costs in a factory. It tells you what you're spending, not whether you're producing profitable products. Business outcomes matter more than token consumption. > Tokens are simply the fuel. The real measure of profitability is how effectively that fuel is converted into measurable business value. ## Three layers of AI profitability Infrastructure efficiency measures cost per token. AI efficiency measures cost per workflow. Business value measures productivity gains and revenue growth. The industry will evolve from tokenomics to business outcomes. ## Supply catches up in 18 to 36 months Power availability is now the pacing item, not just GPU manufacturing. Demand will keep growing as inference, agents, and robotics scale. Commodity GPU pricing will soften, pressuring neoclouds with debt. ## Three phases of market evolution Twenty twenty-six to twenty twenty-seven: supply constrained, premium pricing. Twenty twenty-seven to twenty twenty-nine: new capacity online, commodity pricing, weak neoclouds consolidate. Beyond twenty twenty-nine: efficiency and service differentiation matter most. ## Concentration risk is temporary Today a handful of foundation model labs drive AI infrastructure demand. This mirrors early cloud computing when only internet companies drove adoption. Enterprise demand will eventually diversify and dominate. > We're effectively building the railroads before the towns exist. ## Neoclouds are becoming AI platforms Neoclouds started as GPU providers but now offer orchestration, storage, and MLOps tooling. Next they'll add inference optimization and RAG pipelines. Winners will scale beyond hardware with software and enterprise support. ## Hybrid cloud is the future Enterprises will keep traditional workloads on hyperscalers while running AI training and inference on whichever platform offers best economics and performance. Boundaries between platforms will blur over time. ## Software, not GPUs, wins long term As GPU supply improves, hardware advantage erodes. Neocloud differentiation will come from software, operational excellence, and customer success. The provider that makes AI deployment as simple as spinning up a virtual machine wins. ## Key takeaway Enterprise AI adoption will accelerate when GPU costs drop and companies shift from measuring infrastructure metrics to measuring business outcomes.