The massive, multi-trillion-dollar bet Big Tech has placed on AI infrastructure is about to hit a wall. According to analyst David Cahn, 2026 will be defined by a fundamental split: a “Year of Delays” for data centers and AGI timelines, colliding head-on with the relentless, hyper-efficient acceleration of AI adoption at the application layer.
Google and Meta are fully committed, aggressively positioning themselves for an AI-first future, but the physical reality of scaling compute is proving far slower than the software hype. Cahn notes that while demand for AI CapEx is stronger than ever, the supply chain is showing signs of fatigue and worry about being "left holding the bag."
The first major bottleneck is the semiconductor supply. As Ben Thompson previously detailed, the "TSMC Brake" is real. Companies like TSMC and ASML hold monopolistic positions and cannot be forced to ramp capacity to match the hyperscalers’ demands. TSMC has ramped revenue by 50% since 2022 but CapEx by only 10%. This constraint, especially following major launches like Gemini 3, is expected to become a material factor in 2026, slowing the delivery of crucial chips.
Beyond the chips, the physical construction of data centers, which typically takes two years, is facing industrial bottlenecks. Generators, cooling units, and specialized labor shortages are all poised to push timelines out. If hyperscalers begin warehousing new AI chips instead of installing them, that will be the telltale sign that the era of delays has begun.
The second major delay involves the timeline for Artificial General Intelligence (AGI). For years, Silicon Valley luminaries tossed around "AGI in 2027." That consensus has now progressively walked back. Following recent interviews with key AI researchers, the new, quiet consensus is that the AGI window is now in the 2030s, at the earliest. This update, Cahn argues, will filter outside of the Valley in 2026, raising questions about whether today’s massive CapEx investments will be outdated before AGI ever arrives.