Coverage of NVIDIA GTC conference focuses on the $1 trillion AI infrastructure buildout and competitive dynamics.

The artificial intelligence industry is in the midst of the largest infrastructure buildout since the birth of the cloud, but with a fundamental difference: the AI era is becoming a global infrastructure race centered on securing power, booking semiconductor capacity, locking in memory supply, and deploying massive clusters designed to produce intelligence at scale. What defines this era is the shift from a purely software revolution to something far more capital-intensive — a trillion-dollar supply chain trench war to build the factories that will manufacture intelligence for the next decade. This factory is no longer confined to hyperscale data centers but is beginning to extend outward into what is called the hyperconverged edge, where AI moves closer to where data is created and decisions are made. Companies leading this transformation — including Nvidia, Amazon.com Inc., Microsoft Corp., Google LLC, and Meta Platforms Inc. — are collectively investing hundreds of billions of dollars in AI infrastructure, with some estimates suggesting the next phase of this cycle could approach $1 trillion in capital investment.

An AI factory is fundamentally different from traditional data centers in that it is a vertically integrated system designed to convert raw inputs — power, silicon, memory and data — into outputs such as AI models, inference services, automation and reasoning systems. One economic dynamic peculiar to this cycle is what is termed the GPU appreciation paradox: instead of depreciating rapidly like traditional hardware, the productivity of chips like the widely deployed Nvidia H100 GPU is increasing as the models they serve become more powerful. As frontier models improve, the value generated by the compute serving those models rises, which is why some AI labs are locking in multiyear GPU contracts at roughly $2.40 per hour, well above estimated build costs. AI compute has become the most constrained resource in the digital economy, turning GPUs into something closer to productive capital assets than traditional information technology hardware.

However, GPUs may get the headlines, but the real bottleneck in the AI factory could be memory. Modern AI systems rely heavily on long-context reasoning, meaning models can process huge sequences of text, code and multimodal data, a capability that requires massive amounts of high-bandwidth memory or HBM. Industry data suggest that by 2026, as much as 30% of hyperscaler capital expenditures could go toward memory alone, a shift that is already reshaping semiconductor allocation across the industry, with memory supply increasingly prioritized for AI infrastructure. The semiconductor fabrication capacity itself presents a critical constraint, with datacenter accelerators now consuming a growing share of manufacturing capacity at TSMC. The manufacturing process has two major constraints: front-end capacity in the upstream wafer fabrication stage where advanced logic process nodes produce silicon, and back-end capacity in the mid-end where advanced packaging technologies such as CoWoS integrate chips with high-bandwidth memory and other components. While the back-end packaging stage was the biggest bottleneck last year, the constraint is increasingly shifting toward front-end wafer fabrication capacity.

The ultimate gatekeeper in the semiconductor ecosystem is ASML Holding NV, the Dutch manufacturer of extreme ultraviolet lithography machines required to produce the most advanced chips. Each machine costs more than $350 million, contains hundreds of thousands of components, and relies on a supply chain of thousands of specialized suppliers. More importantly, production is limited, with ASML able to manufacture roughly 70 to 100 extreme ultraviolet tools per year through the end of the decade. That output effectively caps how quickly the world can expand advanced semiconductor production, establishing a fundamental physical constraint on the pace of AI infrastructure scaling that cannot be overcome through software innovation alone.