Some thoughts ahead of Nvidia tonight
Original author: @GavinSBaker
Compiled by: Peggy, BlockBeats
Editor’s Note: After NVIDIA’s earnings report, market attention often focuses on revenue, profit, and guidance ranges. But in this article, author @GavinSBaker seeks to refocus the discussion on a longer-term perspective: what determines NVIDIA’s value is not quarterly figures, but how long AI demand will last and whether compute investments are truly generating sustainable returns.
Drawing on historical experiences of technological cycles, this article examines whether "bubbles and overbuilding" may recur, while highlighting that this AI cycle faces bottlenecks in power and wafer supply, potentially leading to a more restrained expansion pace. On the other hand, GPU rental prices and high utilization rates of older-generation chips provide real-world validation of AI ROI.
The following is the original text:
Here are some personal observations that may be helpful for those following NVIDIA. In my view, the only two core variables worth discussing around this company are the sustainability of demand and the return on investment (ROI) in AI, with the latter being closely tied to the effective lifespan of GPUs.
Sustainability of demand: Will history repeat itself?
Historical experience with technological waves shows that nearly all such cycles have experienced financial bubbles and overexpansion of capacity. Carlota Perez provides a systematic analysis of this in her book *Technological Revolutions and Financial Capital*. She notes that with every technological revolution—whether railroads, radio, or the internet—financial markets tend to identify the long-term potential early, but the ensuing capital frenzy often fuels bubbles (a phenomenon also explainable by Mauboussin’s concept of “collapse in opinion diversity”). These bubbles lead to overbuilding, which in turn causes a temporary decline in demand and ultimately triggers market crashes; meanwhile, the resulting oversupply of foundational technologies paves the way for the “golden age.” The development trajectory of the internet is a classic example.
Therefore, for NVIDIA, the key is not this quarter’s performance or next quarter’s guidance, which are often already fully anticipated by institutional buyers. What truly matters is the sustainability of earnings per share (EPS), not the growth rate for the year.
From the expectations implied by current valuations, the market appears to be signaling a belief that NVIDIA's earnings may be nearing a cyclical peak, driven by concerns over excessive capital expenditure. It is important to emphasize that the market's concern is not a "valuation bubble," but rather a "fundamental bubble"—specifically, the risk of overbuilding fueled by capex. If the market can gain confidence that NVIDIA will sustain a high single-digit revenue CAGR beyond fiscal year 2027, the valuation baseline could find support.
Is this time really different?
“The time is different this time” is often a dangerous assessment. But this AI cycle truly has distinct differences: there are substantial bottlenecks globally in two key dimensions—electricity (watts) and advanced-process wafers—and alleviating these constraints may take several years.
This supply-side hard constraint may, paradoxically, curb excessive capacity expansion. Large-scale cloud providers, if conditions allowed, would theoretically continue to ramp up expansion—but in reality, power and wafer limitations are constraining their pace. Unlike the historical technological revolutions described by Perez, there were no similar supply bottlenecks limiting deployment speed at that time.
Without overbuilding, a crash is unlikely, especially given that current tech stock valuations are not at extreme highs.
Between these two bottlenecks, wafers may be more critical than power. The pace of wafer capacity expansion could become a key variable extending the AI cycle. TSMC’s management has always been known for its caution, emphasizing industry stability and long-term value over short-term aggressive expansion. Without constraints on power and wafers, NVIDIA’s growth over the next 24 months might be faster, but the risk of overbuilding would also rise significantly.
In a sense, supply constraints may be helping to slow and stabilize the entire AI cycle. AI’s heavy reliance on advanced-node wafers may instead be the key factor preventing drastic fluctuations in this cycle.
To realize certain extreme hypothetical scenarios, computing power may need to increase by hundreds or even thousands of times its current level. The time required for this expansion itself provides a buffer for societal adjustment and institutional adaptation.
Historical experience also provides a reference: decades passed after James Watt invented the rotary steam engine before rail systems fully replaced horses. While AI may iterate more quickly, it is still unlikely to overhaul societal structures in an extremely short time.
More importantly, humans achieve “general intelligence” with just 20–30 watts of power. In a world with limited electricity, this efficiency advantage will endure. Therefore, a smoother, more sustained AI cycle may not be bad for society itself.
GPU lifespan and the real ROI of AI
The rental price of GPUs fundamentally reflects the economic value of the token and serves as the core metric of "AI ROI." Theoretically, as higher-performance chips continue to be released, the rental prices of older GPU models should gradually decline, even if AI return on investment remains positive.
However, over the past two months, the rental price of the H100, which has been in service for nearly four years, has risen significantly. This indicates that computing power is generating real and substantial economic value, particularly in agentic AI and code generation scenarios.
Meanwhile, even with the release of Blackwell, the A100, which is six years old, continues to maintain high utilization, and rental prices have not shown significant decline. This strongly suggests that the effective lifespan of GPUs may be at least six years or even longer than the depreciation cycles of most customers.
This has a structural impact: if the residual value is higher than previously expected, the financing cost of GPUs will decrease further. In contrast, ASICs customized for a single model or specific use cases find it difficult to achieve similar lifecycle advantages. In a rapidly evolving environment, specialized chips face higher capital costs and greater difficulty in securing financing.
To some extent, versatility is GPU’s moat. As the functions of prefill and decode are separated and the supporting chip ecosystem gradually takes shape, the computing architecture is evolving from a “single-chip logic” to a “multi-chip collaborative system.” AI infrastructure no longer relies on a single component, but rather on an entire highly integrated system engineering approach.
As prefill and decode become decoupled, the NVIDIA ecosystem may undergo structural adjustments earlier than the TPU ecosystem. Combined with differing design trade-offs among various vendors, the relative cost advantages for customers in inference are shifting.
If some manufacturers previously relied on cost advantages to lower token prices and gain market share, the market behavior will become more rational as this advantage weakens. In the long term, this will positively impact AI ROI, particularly during the transition of compute demand from training to inference.
This shift may be more noteworthy than any quarterly earnings report.
One final lighthearted wish: I hope NVIDIA will bring back superhero names as chip codenames in the future. Surprisingly, the "green team" has never used the name "Banner" (the real name of Marvel’s Hulk).
Click to learn about the open positions at BlockBeats
Welcome to the official BlockBeats community:
Telegram subscription group: https://t.me/theblockbeats
Telegram group: https://t.me/BlockBeats_App
Official Twitter account: https://twitter.com/BlockBeatsAsia