Nvidia Is Flat in 2026. Goldman Says That Makes It Cheap. Who's Right?

TL;DR

• NVDA is up about 5% in 2026, parked near $197. After back-to-back years of face-melting returns, flat feels like a funeral.
• Goldman's take: at a 21.7x forward P/E, near the S&P average and miles below its own five-year average of 72x, Nvidia is cheap. That word, for this stock, is new.
• The bear drip: DeepSeek is building its own AI chip, joining every hyperscaler's custom-silicon program. Death by a thousand in-house chips.
• The stock most likely to be mispriced is the boring one everyone stopped arguing about. NVDA is now that stock.

The Strangest Chart In Tech

Nvidia's 2026: the most dominant company in the most important industry, going sideways.

Sit with the absurdity. The AI buildout accelerated this year: SK Hynix raised $26.5 billion in the biggest foreign IPO in US history to fund memory for Nvidia-centric data centers, the 2026 semiconductor forecast got upgraded to $1.3 trillion, and NAND is physically running out because AI servers eat it. The company at the center of all of it returned 5%.

Meanwhile the multiple did all the work in reverse. Earnings kept growing while the price stood still, compressing NVDA from a 72x five-year-average stock to 21.7x. The market spent 2026 converting Nvidia from a story stock into a value stock without moving the price. Goldman noticed, and called the entry "compelling."

The Bear Case Deserves Its Due

• Custom silicon is real attrition. DeepSeek designing its own chip made headlines (NVDA slipped 2%), but it's one name on a long list: every major hyperscaler has an in-house program aimed at reducing dependence on Nvidia's margins. Each one that ships erodes the monopoly pricing a little more.
• The Kyber scare showed the nerves. A single report claiming the Kyber NVL144 platform would slip to 2028 hit the tape; Nvidia denied it and the stock recovered on the denial. Stocks that swing on roadmap rumors are stocks where confidence is thinner than the fundamentals suggest.
• Prediction markets see more nothing: the highest-probability month-end outcome sits below $192. The crowd has fully internalized "Nvidia does nothing" as the base case.

Why The Boredom Is The Opportunity

Here's the asymmetry. When 72x was the multiple, perfection was priced and every wobble cost 15%. At 21.7x, the market prices Nvidia like a mature industrial: modest growth, margin erosion, no surprises. But the order books, the CUDA moat, and the fact that every custom chip program still buys Nvidia hardware while it tries to replace it all say the erosion is slower than the multiple assumes. Cheap plus dominant plus universally ignored is the setup value investors wait entire cycles for, wearing the ticker everyone already knows.

The Options Angle

• This is the single best LEAPS setup on the board. A flat year means implied volatility has bled out of NVDA options; a compressed multiple means the downside is cushioned. Buy January 2027 calls at or slightly above the money. You're paying deflated vol for a claim on the multiple re-rating even partway back toward historical norms.
• Fund them by selling 30-to-45-day covered calls against long stock if you hold shares. A stock the prediction markets expect to stay under $192 pays you monthly for agreeing with them, while the LEAPS keep the right tail.
• Skip short-dated calls entirely. The re-rating thesis has no scheduled catalyst; it needs an earnings print or two to force the repricing. Give the trade six-plus months of runway or don't take it.

Cheap Or Finished?

Cheap. "Finished" requires believing the custom-silicon wave replaces Nvidia faster than the AI buildout grows the pie, and every data point this year (the $1.3 trillion forecast, the SKHY feeding frenzy, the memory shortage) says the pie is growing faster. The market got bored, not bearish, and boredom prices better than fear. Buy the yawn.