Beyond the Financials, Nvidia's True Risk and Opportunity

Original Title: Some thoughts ahead of Nvidia tonight
Original Author: @GavinSBaker
Translation: Peggy, BlockBeats

Editor's Note: After Nvidia's earnings call, the market's focus is often on revenue, profit, and guidance range. However, author @GavinSBaker in this article attempts to shift the discussion to a longer-term dimension: what determines Nvidia's value is not just one quarter's data, but how long AI demand can sustain and whether the investment in computing power has truly created a sustainable return.

The article starts from historical experience of technology cycles, discussing whether "bubbles and overbuilding" will replay, while pointing out the current AI cycle faces challenges of power and semiconductor supply constraints that may temper the expansion pace. On the other hand, GPU leasing prices and high utilization of older model chips also provide a real-world validation for "AI ROI."

Below is the original text:

Here are some personal observations that may be of reference to those interested in Nvidia. In my view, there are only two core variables truly worth discussing around this company: one is the sustainability of demand, and the other is the return on investment (ROI) in AI, which is closely related to the effective life of GPUs.

Sustainability of Demand: Will History Repeat Itself?

From the historical experience of technological waves, almost all similar cycles have gone through financial bubbles and capacity overexpansion. Carlota Perez's book "Technological Revolutions and Financial Capital" provides a systematic discussion of this. She points out that with each technological revolution, whether it's railways, broadcasting, or the internet, the financial market tends to early recognize its long-term potential, and the ensuing capital enthusiasm often breeds bubbles (which can also be explained by Mauboussin's concept of "collapse of diversity of viewpoints"). Bubbles lead to overbuilding, overbuilding triggers a temporary decline in demand, which then leads to a market crash; and the oversupply of foundational technology ultimately lays the groundwork for a "golden age." The development trajectory of the internet is a typical case.

Therefore, for Nvidia, the key lies not in this quarter's performance or next quarter's guidance, as these are often fully anticipated by buying-side institutions. What truly matters is the sustainability of earnings per share (EPS), not the year-on-year growth rate.

From the implied expectations in the current valuation, the market seems to be expressing a judgment: Nvidia's earnings may be nearing a cyclical peak, with an underlying concern about overexpansion of capital expenditures. It is important to emphasize that the market's concern is not about a "valuation bubble" but a "fundamental bubble," namely the potential risk of overbuilding driven by capex. If the market can gain confidence in Nvidia maintaining a high single-digit revenue compound annual growth rate (CAGR) post the fiscal year 2027, the valuation center may find support.

Is This Time Really Different?

“This time is different” is often a dangerous judgment to make. However, there are indeed unique aspects to this current AI cycle: there are substantial global constraints in two key dimensions, power (watts) and advanced process wafers, and alleviating these constraints may take several years.

This supply-side hard constraint may have actually restrained capacity expansion. Hyperscale cloud providers would theoretically continue to expand aggressively if conditions allowed, but the reality is that power and wafer limitations are restricting their expansion pace. Unlike historical technological revolutions described in Perez's book, there were no similar supply bottlenecks back then to limit deployment speed.

Without overbuilding, a collapse is less likely to occur, especially considering that the overall valuation of tech stocks is not at an extreme high at the moment.

Between these two bottlenecks, wafers may be more critical than power. The pace of wafer capacity control could become a key variable in extending the AI cycle. TSMC's management has always been known for their caution, emphasizing industry steady state and long-term value rather than short-term aggressive expansion. If not for the constraints in power and wafers, NVIDIA's growth in the next 24 months might be faster, but the risk of overbuilding would also significantly increase.

In a way, supply constraints may be contributing to an “AI cycle slowdown steady state.” AI's high dependence on advanced process wafers may actually be a key factor in smoothing out the fluctuations in this cycle.

If we were to entertain some extreme hypothetical scenarios, the scale of computing power may need to increase to hundreds or even thousands of times its current level. The time required for such expansion would in itself provide a buffer for societal adjustment and institutional adaptation.

Historical experience also offers a reference: after James Watt invented the steam engine, it took several decades for the railway system to truly replace horses. The iteration speed of AI may be faster, but it is still not likely to completely restructure societal organization in a very short period of time.

More importantly, achieving “general intelligence” in humans only requires 20–30 watts of power. In a world constrained by power availability, this efficiency advantage will persist in the long term. Therefore, a smoother, more enduring AI cycle may not necessarily be a bad thing for society.

GPU Lifespan and the Real ROI of AI

The rental price of GPUs fundamentally reflects the economic value of tokens and is a core indicator of “AI ROI.” In theory, as higher-performance chips continue to be introduced, the rental price of older GPU models should gradually decrease, even if the AI investment return rate is positive.

However, over the past two months, the H100, which has been in service for nearly four years, has seen a significant increase in rental prices. This means that, especially in the agentic AI and code generation scenarios, computing power is creating real and substantial economic value.

At the same time, even with the introduction of the Blackwell, the A100 from 6 years ago continues to maintain high utilization rates, and rental prices have not shown a significant decline. This strongly suggests that the effective lifespan of GPUs may be at least 6 years, surpassing even the depreciation cycle of most customers.

The impact of this is structural: if the residual value is higher than previously expected, the financing cost of GPUs will further decrease. In contrast, ASICs designed for a single model or specific use case are unlikely to have a similar lifecycle advantage. In a fast-paced iterative environment, the capital cost of specialized chips is higher, making financing more challenging.

To some extent, the universality of GPUs serves as a moat. With the separation of prefill and decode functions and the gradual formation of a complementary chip ecosystem, computing architecture is evolving from "single-chip logic" to a "multi-chip collaborative system." AI infrastructure no longer relies on a single device but rather on a highly integrated system engineering.

With the decoupling of prefill and decode, the NVIDIA ecosystem may undergo structural adjustments earlier than the TPU ecosystem. Coupled with the design trade-offs among different vendors, the relative advantage of customers in the inference cost is changing.

If some vendors previously relied on cost advantages to lower token prices and gain market share, when this advantage diminishes, market behavior will tend toward rationality. In the long run, this will have a positive impact on AI ROI, especially during the transition of computing power demand from training to inference.

This inflection point may be more worthy of attention than any quarterly performance.

One last lighthearted wish: hope that NVIDIA will once again use superheroes as chip codenames in the future. Surprisingly, the "Green Team" has never used the name "Banner" (the real name of the Marvel character Hulk) thus far.

[Original Article Link]