The next wave of AI silicon may pivot from today's compute-heavy architectures to memory-centric ones optimized for inference. This fundamental shift would allow high-performance chips to be produced on older, more accessible 7-14nm manufacturing nodes, disrupting the current dependency on cutting-edge fabs.

The sustainability of the AI infrastructure boom is debated. One view is that GPUs depreciate rapidly in five years, making current spending speculative. The counterargument is that older chips will have a long, valuable life serving less complex models, akin to mainframes, making them a more durable capital investment.

While the industry standard is a six-year depreciation for data center hardware, analyst Dylan Patel warns this is risky for GPUs. Rapid annual performance gains from new models could render older chips economically useless long before they physically fail.

Hyperscalers are extending depreciation schedules for AI hardware. While this may look like "cooking the books" to inflate earnings, it's justified by the reality that even 7-8 year old TPUs and GPUs are still running at 100% utilization for less complex AI tasks, making them valuable for longer and validating the accounting change.

Hyperscalers face a strategic challenge: building massive data centers with current chips (e.g., H100) risks rapid depreciation as far more efficient chips (e.g., GB200) are imminent. This creates a 'pause' as they balance fulfilling current demand against future-proofing their costly infrastructure.

The massive investment in AI infrastructure could be a narrative designed to boost short-term valuations for tech giants, rather than a true long-term necessity. Cheaper, more efficient AI models (like inference) could render this debt-fueled build-out obsolete and financially crippling.

The current AI investment boom is focused on massive infrastructure build-outs. A counterintuitive threat to this trade is not that AI fails, but that it becomes more compute-efficient. This would reduce infrastructure demand, deflating the hardware bubble even as AI proves economically valuable.

Unlike the railroad or fiber optic booms which created assets with multi-decade utility, today's AI infrastructure investment is in chips with a short useful life. Because they become obsolete quickly due to efficiency gains, they're more like perishable goods ('bananas') than permanent infrastructure, changing the long-term value calculation of this capex cycle.

Responding to the AI bubble concern, IBM's CEO notes high GPU failure rates are a design choice for performance. Unlike sunken costs from past bubbles, these "stranded" hardware assets can be detuned to run at lower power, increasing their resilience and extending their useful life for other tasks.