Nvidia’s own FY2028 revenue guidance, delivered this week, points toward roughly $700 billion, nearly $400 billion above what the average sell-side analyst was modeling for the same fiscal year twelve months ago. That gap is not a rounding error. It recalibrates how much weight the market should put on consensus estimates for a company still expanding its addressable compute market. The bigger story here is not the quarter Nvidia just reported. It is how consistently the professionals paid to forecast Nvidia’s business have undershot it.
A year ago, according to MBI Deep Dives, an equity-research newsletter that publishes daily notes on public tech and semiconductor names, the average sell-side consensus for Nvidia’s FY2028 revenue sat near $310 billion. Nvidia’s guidance this week, implying roughly 70% revenue growth for the fiscal year, points to approaching $700 billion instead. Even analysts who spent the intervening months revising their models upward stayed short. MBI Deep Dives put the pre-call consensus at roughly $574 billion, meaning Nvidia’s own guide still beat the market’s updated number by about $125 billion.
A forecasting error that large, repeated in the same direction across a full year of revisions, says less about Nvidia than it does about the value of sell-side consensus itself. A market that has been wrong by $125 billion in one direction is a market whose baseline carries little predictive information, and that cuts both ways. The same mechanism that undershot Nvidia’s growth for a year could just as easily overshoot the next name it covers, or overshoot Nvidia’s own growth if demand catches up to supply.
Nvidia executives told investors on the earnings call that the 70% growth outlook is limited by supply rather than demand, and that revenue could roughly double next year if manufacturing capacity were not the constraint. That is the company’s own characterization of its bottleneck, not an independent projection. MBI Deep Dives argues that, given the tone of the call, 70% growth may function as a floor rather than a ceiling for next year. That reading is the analyst’s extrapolation from Nvidia’s supply-constraint language. It is not a figure Nvidia itself guided to, and readers should hold the two claims apart.
Despite the scale of the miss, Nvidia’s stock has only roughly matched the S&P 500 over the past year and lagged both the Nasdaq 100 and a semiconductor-sector ETF, per KoyFin data cited by MBI Deep Dives. That divergence suggests buy-side investors already discount sell-side Nvidia models more heavily than the models themselves imply. Much of that discount traces to compute concentration: Nvidia disclosed that OpenAI’s commitments total 12 gigawatts, only about 40% of OpenAI’s total chip footprint once AMD, Broadcom, and Amazon deals are counted, while Anthropic’s roughly 3 gigawatt Nvidia commitment is closer to 20% of its total. Both labs are diversifying suppliers even as their overall compute orders grow.
For anyone modeling AI infrastructure spend off analyst consensus, the lesson is not that Nvidia will keep beating estimates forever. It is that sell-side numbers for compute demand have carried a wide, directionally consistent error band for a full year, and treating that consensus as a ceiling rather than a starting point has already cost forecasters $125 billion of accuracy once this year alone.
Figures and analysis per MBI Deep Dives, published August 27, 2026.
