The artificial intelligence boom may still result in a capital bubble, even if the technology itself proves successful.
Such an outcome does not require AI to fail. It only requires supply and capital commitments to exceed monetizable demand. When revenue-producing capacity lags behind deployment, pricing will adjust, and financing will no longer cover the gap. That marks the point when the cycle resets.
Bottlenecks delay the reckoning
The AI supply chain faces constraints in high-bandwidth memory, advanced packaging, network fabric, power, and site readiness. These limitations slow deployment and postpone price discovery—the moment when buyers gain more options and the market determines whether overbuilding has occurred.
For now, those constraints serve two functions. They limit the amount of capacity reaching the market and delay the market-clearing test: the point when the industry learns how much capacity customers will use and what they will pay once supply becomes widely available.
The underlying concept of absorption can be measured by examining whether additional capacity creates enough value and adoption moves quickly enough to absorb it. If system volume expands faster than useful demand and productive utilization, surplus emerges.
The issue extends beyond when more GPUs or memory will ship. It involves when the full system will deliver more revenue-producing capacity than monetizable demand can absorb.
A $1.5 trillion semiconductor market
Global semiconductor revenue neared $800 billion in 2025. A forecast for 2026 projects the market at $1.51 trillion—nearly doubling in a single year.
Nvidia reported $75.2 billion in data-center revenue in its latest fiscal quarter. Broadcom generated $10.8 billion from AI semiconductors, while AMD contributed $5.8 billion in data-center revenue. These figures confirm strong investment in enterprise technology.
Related: Hacking wave targets major financial firms
The forecast’s composition reveals a tension. Over half of the projected 2026 market stems from memory. The revenue growth is driven by two factors: structural AI demand and extraordinary scarcity pricing. This does not reduce demand but indicates revenue is rising faster than physical units and deployment.
If memory pricing stabilizes while physical bit demand and utilization continue growing, the market could shift to a volume-led cycle. However, if prices drop and bit growth stalls, surplus could arrive before demand absorbs it.
Additional HBM supply does not automatically become deployable capacity. The memory must be qualified, stacked, and integrated with accelerators through complex packaging at acceptable yields. A memory stack in inventory remains unproductive until it is packaged, installed, and used for revenue-generating workloads.
Two clocks govern the buildout
AI factories operate on two distinct timelines. The short-cycle IT clock includes GPUs, custom accelerators, HBM, networking, and servers. These components can be ordered, manufactured, and delivered within quarters, with refresh cycles of three to six years.
The long-cycle site clock covers land, buildings, substations, grid connections, cooling, and water infrastructure. These assets require a decade to permit and construct, with no comparable refresh cycle. A data-center building may be physically complete but still waiting for transformers, switchgear, or usable power.
The mismatch between these timelines creates risk. Capital may be committed, hardware allocated, and suppliers report strong orders. Yet the capacity must still be installed, energized, and put to productive use before generating sustainable revenue. Productive utilization involves more than turning on GPUs—it requires running workloads that produce sufficient volume, pricing, and margin to justify the investment.
The timing gap is where bubble risk builds. The concern is not excessive spending but that commitments, hardware deliveries, and supplier revenue outpace the capacity that can be energized and monetized.
Once bottlenecks clear, the market will face its real test: whether demand can absorb the supply surge before pricing and margins collapse. If adoption slows, even robust infrastructure may struggle to generate returns.