NVIDIA (NVDA). Vertiv (VRT) $2.88B This autumn income, orders up 252%. Marvell (MRVL) $8.2B income up 42%. Astera (ALAB) $270.6M income up 92%. Micron (MU) $13.6B income up 57%. Lumentum (LITE) $665.5M income up 65.5%, NVIDIA invested $2B.
As AI information facilities scale previous 100,000 GPUs, the bottleneck shifts from chip provide to bodily infrastructure together with cooling, energy distribution, networking, reminiscence, and optical interconnects.
The analyst who known as NVIDIA in 2010 simply named his high 10 AI shares. Get them right here FREE.
NVIDIA (NASDAQ:NVDA) has not solely turn out to be a trillion-dollar firm as a result of it made one of the best chips, but it surely turned one as a result of it was seemingly the one firm that might ship what the AI business wanted, precisely when it wanted it. This sort of structural dependence is what turns a inventory right into a generational winner. Because of this, the query each investor needs to be asking proper now’s easy: Who could possibly be the subsequent NVIDIA?
Whilst a GPU scarcity helped outline 2023 and 2024, as AI information facilities scale in 2026 and past, the constraints are shifting, and the chips are transport. As a substitute, what’s in brief provide now’s the bodily infrastructure to make these chips helpful. The cooling methods, high-speed networking, reminiscence, and the optical interconnects, simply to call a number of. Hyperscalers will spend lots of of {dollars} in capex this yr, and a rising share of that cash is flowing to firms most buyers have not been paying sufficient consideration to.
The 5 shares under every sit at a chokepoint within the AI provide chain that arguably did not exist at this scale three years in the past. Better of all, these are usually not speculative performs, as they’ve accelerating income, document backlogs, and direct relationships with NVIDIA, however the market hasn’t absolutely priced in what these positions are price.
READ: The analyst who known as NVIDIA in 2010 simply named his high 10 AI shares
NVIDIA’s GPUs are ineffective with out the infrastructure to energy them, cool them, feed them information, and transfer alerts between them on the pace AI calls for. As clusters scale previous 100,000 GPUs, each a kind of capabilities turns into a possible bottleneck, and the businesses that may assist resolve it will turn out to be more and more indispensable. That is the setup.
If it is NVIDIA transport the chips, it is Vertiv (NYSE:VRT) that’s maintaining them alive as the corporate that’s manufacturing the facility distribution and thermal administration methods that each AI information heart requires earlier than a single GPU goes on-line. As these racks of GPUs log on and push previous 100 kilowatts, their liquid cooling methods have shifted from non-obligatory to necessary.
The corporate’s This autumn 2025 income indicated it hit $2.88 billion with natural orders surging 252%, and full-year income reaching $10.23 billion, all whereas its backlog stands at $15 billion, up 109% year-over-year, with a This autumn book-to-bill ratio of two.9x.
The NVIDIA parallel is much more direct within the sense that simply because the GPU provide could not maintain tempo with demand in 2023 and 2024, cooling and energy infrastructure are actually the binding constraint on how briskly information facilities can deploy AI capability. Vertiv is guiding $13.25 billion to $13.75 billion in 2026 income, which might crush Wall Road’s prior $12.4 billion estimate.
NVIDIA owns computing, however its Marvell Expertise (NASDAQ:MRVL) owns connectivity as its AI clusters scale to lots of of hundreds of GPUs, the pace at which information strikes between chips turns into simply as crucial because the chips themselves. Marvell’s buyer AI ASICs, optical DSPs, and 1.6T interconnect options are the networking material which have helped maintain a few of these clusters collectively.
So far as the numbers go, fiscal 2026 income hit a document $8.2 billion, up 42% year-over-year, with information heart merchandise accounting for 74% of complete gross sales and non-GAAP EPS rising 81%.
The following NVIDIA case with Marvell examines the corporate’s customized silicon enterprise, which grew from close to zero to $1.5 billion in annual income in a single yr. The corporate has 18 design wins with hyperscalers, together with Microsoft and Amazon, and it not too long ago acquired Celestial AI for $3.25 billion to convey photonic interconnect know-how in-house. In Marvell’s case, each main cloud firm is more likely to want its chips to hook up with its AI infrastructure, it is not simply the provider, but in addition a platform.
Each rack of NVIDIA GPUs wants Astera Labs (NASDAQ:ALAB) retimers to perform, and the corporate’s PCI/CXL good DSPs and cable modules guarantee information flows between elements with any type of sign degradation. It is this efficiency that has helped make Astera the connective tissue contained in the GPU rack itself.
Given its crucial nature, it is no shock that Astera’s This autumn income hit $270.6 million, up 92% year-over-year, with 75.7% gross margins heading towards 74% in Q1 as a result of a $6.5 billion Amazon warrant settlement and higher-margin product combine shift tied to Taurus mannequin enlargement.
The inventory itself is down practically 60% from its 2025 peak because the Amazon partnership launched an estimated 200 foundation level quarterly margin drag beginning in Q2, however the structural place stays properly intact regardless of any near-term profitability issues. Astera can be increasing shipments to further hyperscalers, and its Scorpio material targets a connectivity market estimated to achieve $25 billion over the subsequent 5 years.
A $20 billion market cap and with income that’s practically doubling yearly, Astera has the profile of a vital AI infrastructure title.
AI servers require roughly thrice the reminiscence of ordinary servers, and the specialised high-bandwidth reminiscence that feeds NVIDIA’s GPUs is in a structural scarcity that Micron (NASDAQ:MU) is uniquely positioned to make the most of. Micron is the one American HBM producer, and its Q1 fiscal 2026 income got here in at $13.6 billion, up 57% year-over-year, with non-GAAP EPS of $4.78, beating estimates by over 20%. The kicker is that the corporate’s complete 2026 HBM provide is already bought out, together with its next-generation HBM4.
The valuation is the place the NVIDIA comparability turns into compelling as the corporate is guiding Q2 income of $18.7 billion with an EPS of $8.42, a quantity that represents 440% earnings development, but it trades at roughly 9x ahead earnings.
Comparable AI infrastructure names commerce at 25 to 30 instances, all whereas the HBM market is projected to develop from $35 billion in 2025 to $100 billion by 2028, and Micron can presently meet solely 50% to two-thirds of the demand from its key clients. NVIDIA goes to wish Micron’s HBM3E for its Blackwell platform and can want HBM4 for the whole lot that comes subsequent.
With out Lumentum’s (NASDAQ:LITE) lasers, NVIDIA’s scale-out AI architectures is not going to work, and the corporate makes the optical elements and co-packaged optics that carry information throughout the huge clusters NVIDIA’s platforms require.
On March 2, 2026, NVIDIA validated this dependency straight by investing $2 billion in Lumentum, together with multibillion-dollar buy commitments for laser elements. This is not a partnership announcement, as a substitute, that is NVIDIA locking down a provide chain it might probably’t construct with out.
Wanting on the numbers, Lumentum’s fiscal Q2 2026 income reached $665.5 million, up 65.5% year-over-year, and the corporate swung from a web lack of $78.2 million to a web earnings of $1.1 million. Higher but, Q3 steerage requires $780 million to $830 million in income, whereas the optical backlog exceeds $400 million, and analysts mission fiscal 2026 income of roughly $2.6 billion. The inventory has surged over 900% prior to now yr, so the valuation is the clear danger, however with NVIDIA committing to a multiyear buy, it is telling you whether or not Lumentum is in a robust place.
Wall Road is pouring billions into AI, however most buyers are shopping for the mistaken shares. The analyst who first recognized NVIDIA as a purchase again in 2010 — earlier than its 28,000% run — has simply pinpointed 10 new AI firms he believes might ship outsized returns from right here. One dominates a $100 billion tools market. One other is fixing the only greatest bottleneck holding again AI information facilities. A 3rd is a pure-play on an optical networking market set to quadruple. Most buyers have not heard of half these names. Get the free listing of all 10 shares right here.
