‘Everything about the AI bubble breaks,' tech expert claims 75% of chips Nvidia sold remain unused

Nvidia is facing renewed scrutiny over Blackwell GPU shipment claims, with critics arguing that the reported 6 million GPUs may overcount and that even 3 million actual units could be overstated. The article centers on skepticism from AI bear Ed Zitron rather than new operational data, making the impact primarily reputational and limited in immediate market significance.

Analysis

This is less about one disputed shipment metric and more about whether Nvidia’s demand narrative is starting to outrun verifiability. In hardware cycles, credibility erosion usually shows up first in the supplier stack: channel partners, ODMs, and foundry/utilization commentary will start to diverge from headline bookings before the stock fully re-rates. If buyers sense that end-demand is being inferred from accounting-friendly unit math, the multiple compression can happen fast because the market pays a premium for Nvidia's scarcity premium and flawless execution premium, not just growth.

The second-order risk is that a perception issue can cascade into a forecast issue for the entire AI capex complex. If customers, hyperscalers, or regulators begin to question shipment quality or installed-base claims, that can slow purchasing decisions by one to two quarters even if underlying AI spending remains intact. The biggest loser is not necessarily Nvidia's revenue next quarter; it's the visibility of the next four quarters, which matters more for a name priced on forward confidence than on current earnings.

Contrarian angle: this may be more governance optics than fundamental damage, and the market may already be conditioned to dismiss hyperbolic management messaging. The bigger tell will be whether gross margins, inventory turns, and backlog conversion stay clean; if they do, the controversy could fade as noise. But if the company leans into ever-looser definitions to defend growth optics, investors may begin applying a persistent credibility discount that lasts for multiple reporting cycles.

From a timing perspective, this is a days-to-weeks sentiment trade unless corroborated by weaker supply-chain data over the next 1-2 quarters. If the stock holds up despite repeated credibility questions, that would suggest the tape still prioritizes AI scarcity over governance concerns; if it breaks technical support on rising skepticism, the downside can accelerate because crowded long positioning amplifies de-risking. The cleanest setup is to express the thesis with defined risk rather than outright conviction until the next earnings or channel-check cycle.

Key Decisions for Investors

• Short NVDA into strength on any post-headline bounce; use a 2-6 week horizon and cover on evidence that gross margin and backlog commentary remain unchanged. Risk/reward is attractive if the market starts pricing a credibility discount, but size modestly because AI momentum can overwhelm near-term narrative damage.
• Buy NVDA put spreads 1-2 months out to express a controlled downside view on multiple compression, not an earnings collapse. Prefer spreads over outright puts because the stock can remain expensive longer than skepticism persists.
• Pair trade: short NVDA / long a basket of AI infrastructure beneficiaries with cleaner optics and less narrative risk, such as semiconductor equipment or networking names, over the next quarter. This isolates the governance premium at NVDA while keeping exposure to continued AI capex.
• If you want lower-beta exposure, wait for confirmation from upcoming results or channel data before adding to shorts; the catalyst window is next earnings plus any supply-chain commentary over the next 1-2 quarters. A failed rebuttal from management would be the signal to press the trade.
• For existing longs, trim 20-30% into any rally until the company provides cleaner disclosure on shipment/accounting methodology. The main risk is not immediate revenue loss but a sustained credibility haircut that can persist for several quarters.