Original Title: Some thoughts ahead of Nvidia tonight
Original Author: @GavinSBaker
Translation: Peggy, BlockBeats
Editor's note: After Nvidia's earnings report is released, the market's focus often centers on revenue, profit, and guidance ranges. However, the author @GavinSBaker attempts to pull the discussion back to a longer-term perspective: What determines Nvidia's value is not quarterly data, but how long AI demand can be sustained and whether computing power investments truly create sustainable returns.
The article starts from historical experiences of technological cycles, discussing whether "bubbles and overbuilding" will recur while pointing out that this round of the AI cycle faces bottlenecks in power and wafer supply, which may make the pace of expansion more restrained. On the other hand, high rental prices for GPUs and high utilization of older model chips provide a realistic validation for "AI ROI".
Below is the original text:
Here are some personal observations that may be of reference to those focused on Nvidia. In my view, there are only two core variables around this company that are truly worth discussing: one is the sustainability of demand, and the other is the return on investment (ROI) in AI, which is closely related to the effective lifespan of GPUs.
Sustainability of Demand: Will History Repeat Itself?
From the historical experience of technological waves, almost all similar cycles have experienced financial bubbles and excessive capacity expansion. Carlota Perez has systematically discussed this in "Technological Revolutions and Financial Capital." She points out that with every technological revolution, whether it be railroads, broadcasting, or the internet, financial markets tend to recognize its long-term potential early on, and the ensuing capital frenzy often gives rise to bubbles (which can also be explained by Mauboussin's concept of "collapse of diversity of views"). Bubbles lead to overbuilding, overbuilding causes a decline in cyclical demand, which in turn leads to market crashes; and an oversupply of the underlying technology ultimately lays the groundwork for a "golden age." The trajectory of the internet's development is a typical case.
Therefore, for Nvidia, the key is not in this quarter's performance or next quarter's guidance, which are often fully anticipated by buyer institutions. What is truly important is the sustainability of earnings per share (EPS), rather than the growth rate for the year.
From the expectations implied in the current valuation, the market seems to be expressing a judgment: Nvidia's profits may be approaching a cyclical peak, underpinned by concerns over excessive capital expenditure expansion. It should be emphasized that the market's worries are not about a "valuation bubble," but rather a "fundamental bubble," namely the potential risk of overbuilding driven by capex. If the market can build confidence in Nvidia maintaining a high single-digit revenue compound annual growth rate (CAGR) after fiscal year 2027, the valuation center may receive support.
Is This Time Really Different?
"This time is different" is often a dangerous judgment. However, this current AI cycle does have differences: there are substantial bottlenecks globally in two key dimensions: power (watts) and advanced process wafers, and the relief of these constraints may take years.
This hard constraint on the supply side may, in fact, suppress excessive capacity expansion. Hyperscale cloud providers, if conditions permit, would theoretically continue to expand, but in reality, power and wafer limitations restrict their pace of expansion. Unlike the historical technological revolutions described in Perez's book, there were no similar supply bottlenecks at that time to limit deployment speeds.
Without overbuilding, a crash is difficult to occur, especially considering that the overall valuation of tech stocks is not at an extreme high.
Among these two bottlenecks, wafers may be more critical than power. The rhythm of wafer capacity control may become an important variable for prolonging the AI cycle. TSMC's management is known for being prudent; they emphasize industry stability and long-term value rather than short-term aggressive expansion. Without the constraints of power and wafers, Nvidia's growth over the next 24 months might be faster, but the risk of excessive building that comes with it would also significantly increase.
In a sense, these supply constraints may be providing a "deceleration stability" for the entire AI cycle. AI's high dependence on advanced process wafers may, conversely, become a key factor avoiding severe fluctuations in this round of cycles.
If some extreme hypothetical scenarios are to be realized, the scale of computing power may need to increase hundreds or even thousands of times the current levels. The time needed for this expansion itself provides a buffer for social adjustments and institutional adaptations.
Historical experience also offers a reference: After James Watt invented the rotary steam engine, it took several decades for the railway system to truly replace horses. The iteration speed of AI may be faster, but it still won't complete the restructuring of social structures in a very short time.
More importantly, humans need only 20-30 watts of power to achieve "general intelligence." In a world with limited power, this efficiency advantage will exist for the long term. Therefore, a smoother and more enduring AI cycle may not be a bad thing for society itself.
GPU Lifespan and Real ROI of AI
The rental price of GPUs essentially reflects the economic value of tokens and is a core metric for "AI ROI." Theoretically, as higher-performance chips continue to be launched, the rental price of older model GPUs should gradually decline, even if the AI return on investment is positive.
However, in the past two months, the rental price of the H100, in service for nearly four years, has seen a significant increase. This means that, especially in the context of agentic AI and code generation scenarios, computing power is creating real and substantial economic value.
Meanwhile, even with the introduction of Blackwell, the A100 from six years ago still maintains high utilization, and rental prices have not shown significant looseness. This strongly suggests that the effective lifespan of GPUs may be at least six years or more, even exceeding the depreciation cycles of most customers.
The impact of this is structural: if the residual value exceeds previous expectations, the financing costs of GPUs will further decrease. In contrast, ASICs customized for a single model or specific use are difficult to have similar lifecycle advantages. In a rapidly iterating environment, the capital costs of specialized chips are higher, and financing is more challenging.
To some extent, versatility is the moat of GPUs. With the functional disaggregation of prefill and decode, as well as a gradually formed supporting chip ecosystem, the architecture of computing power is evolving from a "single chip logic" to a "multi-chip collaborative system." AI infrastructure no longer relies on a single device but is instead a highly coupled system engineering.
With the decoupling of prefill and decode, Nvidia's ecosystem may complete structural adjustments even earlier than the TPU ecosystem. Coupled with the varying design trade-offs among different manufacturers, customers' relative advantages in inference cost are changing.
If some manufacturers previously relied on cost advantages to reduce token prices to gain market share, then when this advantage weakens, market behavior will tend to rationality. In the long run, this will positively impact AI ROI, especially during the phase of transitioning computing power demand from training to inference.
This turnaround may be more noteworthy than any quarterly performance.
A final lighthearted wish: I hope Nvidia will resume using superheroes as chip code names in the future. Surprisingly, the "green camp" has yet to use the name "Banner" (the real name of the Marvel character Hulk).
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
