To The Moon

Has Nvidia Been Punished Too Harshly at 24x Forward P/E?

With “AI bubble” fears still in the air, $Alphabet(GOOG)$   has emerged as the new tech market star on the back of its Gemini 3 model and full-stack AI strategy. The stock is up over 14% this month and about 70% year to date, leading the M7 and putting Alphabet within reach of a $4 trillion market cap. 

In contrast, the surge of Google’s AI chips has coincided with a sharp selloff in $NVIDIA(NVDA)$   , wiping about $1 trillion off its market value in under a month and dragging other GPU names like $Advanced Micro Devices(AMD)$   lower as well.

The core question for investors is simple: in this wave of TPU enthusiasm, has Nvidia actually been oversold, or is the market just taking some excess optimism out of the story?

What the Market Is Afraid Of

The recent de-rating of Nvidia boils down to three fears:

1. Substitution risk from in-house chips 

Cloud giants are rolling out their own accelerators—Google's TPU line, AWS Trainium and Inferentia, Microsoft's Maia and Cobalt. The worry is that, over time, these efforts sharply reduce the hyperscalers' dependence on Nvidia GPUs.

2. Cost and efficiency gaps 

In specific workloads, Google claims that its latest TPUs offer much higher performance per watt and lower dollar-per-compute than comparable GPUs. If you're a CFO staring at billions in AI capex, a 30–50% cost difference is impossible to ignore.

3. “From must-have to optional” narrative shift 

Gemini being trained end-to-end on TPUs has been framed as proof that frontier models no longer need Nvidia. That feeds the fear that AI demand may peak sooner than expected, and that Nvidia's golden era of monopoly-like economics is behind it.

What TPUs Really Are: A Specialized Tool, Not a Universal Replacement

The limitations of ASIC are also clear:

1. Closed ecosystem

TPUs are only available on Google Cloud. Using them at scale inherently locks customers into GCP, which many enterprises are reluctant to do. Multi-cloud strategies are hard to reconcile with a chip that lives in exactly one cloud.

2 Weaker software and developer moat

CUDA has had more than a decade to become the “lingua franca” of accelerated computing. It runs everywhere: on consumer GPUs, on-prem servers, every major cloud, and in countless research environments. Tooling, libraries, and community are all built around it.

TPU tooling has improved, but the ecosystem remains smaller and more fragmented. Migrating mature CUDA code to TPU is non-trivial and often not worth the engineering cost.

3. Narrower applicability

Most of the “TPUs are cheaper and faster” claims are benchmarked on a limited set of deep-learning tasks. For broader use cases—traditional HPC, scientific computing, graphics, mixed workloads—GPUs remain far more flexible and better supported.

Demand Side: Market Share Loss ≠ Revenue Peak

Even if we assume that TPUs and other in-house ASICs gain meaningful share at the big clouds, it doesn't automatically follow that Nvidia’s revenue is peaking.

A few key points:

– Nvidia still commands a very high share of the AI accelerator market—often estimated around the 80–90% range today.

– Hyperscalers may shift part of their incremental spend to in-house chips, but they are not walking away from GPUs. Instead, their roadmaps increasingly show coexistence: custom silicon for certain standardized workloads, GPUs for flexibility and peak performance.

– Outside the hyperscalers, most enterprises, governments, and research institutions do not have the scale, talent, or risk appetite to design their own chips. For them, Nvidia’s full-stack solution—GPU, networking, software—is the default option.

The more important variable is overall AI compute demand. If total AI capex continues to compound at a high rate, Nvidia’s unit shipments and revenue can still grow even if its market share falls from, say, 90% to 60–70%. The pie is expanding fast; losing a few slices does not mean the bakery shuts down.

So TPU enthusiasm primarily affects who gets how much of the incremental pie, not whether the pie exists. 

Valuation: At 24x Forward Earnings, the Odds Look Reasonable

During the peak of AI euphoria, Nvidia's forward P/E multiple blasted into the 50–70x zone—levels that effectively assumed years of flawless execution and near-monopoly economics. After the recent sell-off and earnings resets, that multiple has compressed dramatically.

Nvidia now trades around 24x forward earnings, versus roughly 22x for the S&P 500 and high-20s for some other mega-cap tech names. At roughly this valuation, however, the market seems to have largely priced in the market-share and growth impact from ASIC competition—leaving a more attractive risk-reward profile for new bets on Nvidia.

@TigerStars  @CaptainTiger  @TigerWire  @Daily_Discussion  @Tiger_chat  @Tiger_comments  @MillionaireTiger