The Infrastructure Shift: Why Your AI Assets Are No Longer Depreciating

The arrival of Artificial General Intelligence, signaled by GPT-6 Astra, changes how we must value computing infrastructure. We are moving from an era where hardware is a cost center destined for the scrap heap to one where it acts as a durable, revenue-generating asset. This transition helps organizations that stop viewing GPUs through the lens of traditional depreciation schedules and start treating them as fungible, high-yield capital. For investors and operators, the implication is clear: the market is mispricing the longevity and utility of AI hardware, creating a window to capture value before the rest of the sector catches up to this economic reality.

The Revaluation of Hardware as Capital

The conventional view that hardware is a rapidly depreciating asset is failing in the face of current AI demand. When NVIDIA CEO Jensen Huang describes their computing infrastructure as "fungible, durable and highly rentable," he is describing a fundamental change in the economics of the system.

We see this in the rental market for H100 GPUs. Despite being three years old, their rental prices have climbed 22% in a single month. This defies standard accounting logic, where such assets should be losing value. The system is responding to a scarcity of compute power that nullifies the expected depreciation curve.

"NVIDIA Compute is fungible, durable and highly rentable. It is a productive revenue-generating asset."

-- Jensen Huang

Second-Order Effects of GLP-1 Adoption

The rise of GLP-1 medicines like Ozempic and Wegovy is forcing a structural pivot in the consumer staples sector. PepsiCo’s shift from shelf-stable snacks to refrigerated, protein-rich, and fresh food categories is a defensive move against changing consumer biology.

These drugs cause a decline in demand for traditional high-calorie, salty snacks. The downstream effect is a forced migration of capital within the grocery store. Companies that fail to track this shift in consumer behavior are betting on a demographic that is shrinking or changing its habits. PepsiCo’s strategy of extending established brands into fresh categories, such as the Tostita’s Guacamole launch, is an attempt to capture the diverted spending on protein and fiber before competitors secure that shelf space.

The Systemic Breakout in Memory

Goldman Sachs’ observation regarding memory stocks reveals a classic technical pattern: a sector emerging from a long consolidation phase. While the broader AI trade has been volatile, the technical alignment of companies like Micron and Sandisk suggests that the market is beginning to price in the next phase of infrastructure demand.

"The potential breakouts come as the broader AI trade enters a more constructive setup, with the lighter investor positioning, lower implied volatility, and a series of potential catalysts."

-- Lee Coppersmith

In systemic shifts, the picks and shovels often follow the intelligence layer. As models like GPT-6 Astra saturate frontier benchmarks, the demand for the memory required to sustain these models will likely compound, creating a secondary wave of value that is only now beginning to emerge from its summer doldrums.

Key Action Items

• Re-evaluate Hardware Depreciation: Audit your technology assets. If you are treating GPUs or high-end compute hardware as standard three-year depreciating assets, adjust your models to account for their current rentable value. (Immediate)
• Monitor GLP-1 Impact on Consumer Portfolios: If you hold positions in traditional packaged food companies, assess their exposure to the fresh and refrigerated transition. Companies sticking purely to shelf-stable snacks face long-term volume erosion. (Next 6-12 months)
• Track Memory Sector Breakouts: Watch for sustained volume in memory-focused ETFs and individual semiconductor stocks. The current consolidation break is an early-stage signal that may precede broader infrastructure scaling. (Over the next quarter)
• Shift from Product to Infrastructure Thinking: Whether in software or hardware, look for assets that are fungible and rentable. The ability to pivot compute power to different tasks is the new moat. (12-18 months)
• Prepare for Frontier Performance Shifts: With GPT-6 Astra hitting 99.9% on benchmarks, prepare for a rapid acceleration in automated problem-solving within your own organization. If your workflows have not integrated these capabilities, you are already falling behind the efficiency curve. (Next 6 months)