Is NVIDIA still a good stock to buy?
The business is growing at a rate almost no company its size has managed. The share price has gone nowhere this year. Both facts are true, and the gap between them is the entire argument — here it is, on both sides.
The short answer
Nobody can tell you whether to buy it, and anyone who does is guessing. What can be set out is the actual case on both sides, and the specific things that decide it. The core tension: NVIDIA's business is still growing at a rate almost no company of its size has managed, while its share price has gone nowhere this year because investors are questioning whether the customers funding that growth can keep spending.
The case for
The moat is software, not silicon
CUDA is the layer a decade of machine learning work has been written against. Switching to a competitor means rewriting that work, which is why AMD selling a comparable accelerator hasn't translated into a comparable share of the market. Chip leads erode; ecosystems are stickier. See what an economic moat is.
The growth is real, not projected
65% revenue growth at a company of this size is extraordinary and it's in the accounts, not in a forecast. Hyperscale customers have continued raising their capital spending outlooks.
Capital light by chip standards
NVIDIA doesn't own fabs. Manufacturing is outsourced, which keeps capital intensity far below a memory maker or Intel and means free cash flow converts well.
The case against
Customer concentration
A handful of hyperscalers account for the bulk of demand growth, and their capital budgets are discretionary. They can be cut in a quarter. This is the single biggest risk and it isn't a technology risk at all.
Customers are becoming competitors
Google has TPUs, Amazon has Trainium and Inferentia, Microsoft is developing its own accelerators. Every custom chip a hyperscaler deploys is an NVIDIA sale that doesn't happen, and these are the same companies driving the demand.
The circular financing question
Investors have started asking whether AI companies are effectively financing each other's spending — chipmakers with financial ties to the customers buying their chips. If some demand is a closed loop rather than end-customer need, the growth rate is less durable than it looks. This concern knocked nearly 5% off the shares in a single session in late July.
The new debt
June's $25bn bond issue adds roughly $1.2bn of annual pre-tax interest and grows the capital base by $25bn, which mechanically lowers return on invested capital until that money earns more than its cost.
It's a cyclical, and cyclicals mislead
Semiconductors look cheapest on price-to-earnings exactly when earnings are at a cyclical peak. See why a chip stock halving doesn't make it cheap.
The four checks that actually decide it
Normalised earnings
Average operating profit across a full cycle, not this year's. If the current year is more than roughly double the average, you're looking at peak earnings.
What share of revenue depends on continued hyperscaler capex
Work out the number. That's the thesis, stated honestly.
Inventory against revenue
Rising inventory days ahead of a downturn is the classic tell in this industry.
Your own margin of safety
Estimate what the business is worth on assumptions you'd defend in a bad year, then insist on a discount to it. See how to calculate intrinsic value.
The honest conclusion
NVIDIA is an exceptional business at a price that assumes it stays exceptional. That's not the same as a bargain, and it isn't the same as a bubble either. Both camps are arguing about one unknowable: whether AI capital spending sustains at current levels. If it does, today's price will look reasonable. If it slows, a company with 65% growth priced for continuation has a long way to fall.
Put a number on it
Oak Growth calculates a discounted cash flow intrinsic value for NVIDIA and roughly 1,000 other companies, and shows the margin of safety against today's price.
Explore Oak GrowthCommon questions
Is NVIDIA still a good stock to buy?
That depends on a single unknowable: whether hyperscaler AI capital spending sustains at current levels. The business is growing at an extraordinary rate and has a genuine software moat in CUDA, but the price assumes continuation, and a handful of customers who are also building their own chips drive most of the demand.
Why has NVIDIA stock not gone up in 2026?
Shares have risen only around 2% in 2026 against roughly 7% for the S&P 500, despite strong results. Investors have been repricing the whole AI supply chain over questions about whether capital spending is sustainable, and over concerns that some demand may be circular — AI companies financing each other's purchases.
What are the biggest risks to NVIDIA?
Customer concentration is the largest: a few hyperscalers drive most demand growth and their capital budgets are discretionary. Those same customers are developing their own accelerators, which turns buyers into competitors. Semiconductors are also deeply cyclical, so current earnings may represent a peak rather than a run rate.
Is NVIDIA overvalued?
It depends entirely on what earnings you consider normal. Valued on current earnings the multiple looks manageable; valued on average earnings across a full semiconductor cycle it looks far more demanding. That distinction, rather than any single ratio, is the whole argument.
Also see: NVIDIA (NVDA) full analysis → · Are semiconductor stocks undervalued? → · Are AI stocks overvalued? →