Ask an AI assistant to plan your week, and somewhere a rack of servers draws power to answer you. Multiply that by hundreds of millions of people, and the bill gets big fast.
For the past three years, the AI boom’s bottleneck was chips. Now it’s finding enough electricity, land and buildings to plug them into.
Nvidia (NVDA) has been the biggest winner of that boom. Revenue jumped 106% to $96.2 billion in its fiscal second quarter, with data center sales up 117% to $89 billion, according to the company’s earnings release.
“AI has reached its inflection point. It’s doing useful work,” CEO Jensen Huang said in the release.
Still, investors have spent the fall worrying about everything around the chip: power shortages, memory costs and how AI customers will pay for it all. Those fears have weighed on sentiment, even with the stock trading near its 52-week high.
According to a Morgan Stanley research report shared with me, those same constraints may end up playing to Nvidia’s strengths.
Morgan Stanley reinstates Nvidia as its top chip pick
Morgan Stanley recently hosted Huang, CFO Colette Kress and investor relations head Toshiya Hari for investor meetings in New York and Boston. Analyst Joseph Moore came away more convinced and put Nvidia back at the top of his list.
“On our road show this week, NVDA’s CEO highlighted that recent developments in land/power/shell, financing, component constraints, and agentic all play to their strengths; we agree and are reinstating the stock as our Top Pick in semis,” wrote Moore in the Oct. 2 note.
Read more: Why Nvidia & Microsoft are still worth watching as Treasury yields rise
Moore rates the stock Overweight with a $300 price target, about 30% above its $230.86 close on Oct. 1. Nvidia has since climbed to $237.47 at the Oct. 7 close, according to FinancialContent data, which still leaves roughly 26% upside to the target.
Moore made the same call earlier this year. He named Nvidia his top pick in March, TheStreet reported, a move he now says came at least three months too early, and the label expired after six months under Morgan Stanley’s rules.
He also argues the stock is cheap for its growth, at about 15 times fiscal 2028 earnings. His $300 target is roughly 20 times his calendar 2027 earnings estimate of $15.01 a share.
Morgan Stanley makes Nvidia its top chip pick again, sets $300 target.
Data center power shortages tilt the AI chip race
The power squeeze is real. Energy intelligence firm Currence estimates 30% to 50% of large data center capacity expected online in 2026 will be delayed, with electricity a primary obstacle, Network World reported.
Only about five gigawatts of the roughly 16 gigawatts scheduled to start in 2026 was actually under construction, Currence found.
Nvidia’s answer is to squeeze more out of every watt. Its September investor presentation shows revenue per gigawatt rising from $18 billion with Hopper to $25 billion with Blackwell and $40 billion with Rubin.
More Artificial Intelligence:
- Trump’s ‘Super Intelligence’ order rolls through the tech world. But will it last?
- Meta Muse gives AMD and Intel investors a reason to cheer
- Microsoft’s $665 target hinges on a new AI advantage
Moore said Nvidia’s slides put Feynman, its 2028 architecture, well north of $50 billion per gigawatt. Rivals pitch their lower cost per gigawatt as an advantage, and he expects that debate to get louder.
“But if the biggest constraint becomes land/power/shell, the highest tokens per Gigawatt win, all else being equal,” wrote Moore.
Given that backdrop, it’s easy to understand why Moore likes the setup. He wrote that the biggest buyers of cheaper custom chips are the same companies with the biggest data center power problems, and his industry contacts say Nvidia produces many times more tokens per gigawatt than peers.
Nvidia’s own guidance points the same way. Kress said supply will “remain a bottleneck” at least through fiscal 2028, CFO Dive reported.
Amazon and SpaceX carry the case for faster AI growth
On its August earnings call, Nvidia gave a preliminary forecast for 70% revenue growth in fiscal 2028, CFO Dive reported. Moore calls that a supply-constrained number, with underlying demand closer to 100% growth.
Two customers do most of the lifting in his math. Amazon plans to add 2 million more Nvidia GPUs across its cloud in 2027 and 2028, the company announced in August, and Morgan Stanley estimates Amazon’s Nvidia spending will climb to $78 billion in 2027 from about $40 billion this year.
“Customers want the freedom to choose the best tools for their AI workloads,” AWS CEO Matt Garman said in the announcement.
Add SpaceX’s buildout, and Moore estimates the two customers explain about 72% of the
“Conservative forecasts for those customers’ contribution to Nvidia’s 75% datacenter revenue guide implies about 20% growth for the remaining customers in Nvidia’s hyperscale segment,” wrote Moore, well below the 39% capex growth Wall Street expects from that group.
Much will likely depend, however, on those two buyers staying on schedule. In my view, a forecast that leans this hard on SpaceX and Amazon is a low bar for everyone else, but it’s also a concentration risk if either one slows down.
Margins are the other swing factor. Nvidia guided fiscal 2028 gross margin to 72% to 73% as memory costs climb, TIKR reported, and Moore sees that range as a floor because Nvidia can raise prices and trim memory content to offset higher costs.
Moore’s base case doesn’t need investors to fall back in love with AI stocks to work.
“There’s opportunity for multiple expansion if enthusiasm returns to AI but multiple expansion is also not required for the stock to work as numbers continue to go higher,” wrote Moore.
