The $725 Billion AI Capex Cycle Has 3 Bottlenecks: Power, Memory, and Optical Bandwidth. 3 Stocks Poised to Win Big.

Like any other new industry, the artificial intelligence (AI) business continues to evolve as it grows. It's getting better -- and more cost-effective -- by figuring out where it's deficient. And right now, its biggest roadblocks are a lack of memory chips, a strained supply of power, and networking solutions without nearly enough throughput capacity.

It's not afraid to spend big money shoring up these problems either. Given recent spending outlooks from top names in the AI infrastructure business, such as Microsoft, Amazon, and Alphabet's Google, Goldman Sachs now expects $765 billion worth of AI infrastructure investments to be made this year alone. The lion's share of these are outlays likely to be directed at the three aforementioned bottlenecks.

And this raises the question: Which companies are best positioned to benefit from this anticipated capital spending? Here's a closer look at a good guess for each category.

Image source: Getty Images.

Power: GE Vernova

Ordinary utility companies are obviously beneficiaries of the soaring demand for electricity related to the proliferation of AI data centers. The International Energy Agency expects data centers' total power consumption to roughly double between now and 2030, to 945 TWh (terawatt hours). That's enough electricity to support a few dozen major cities.

The only problem? Most utility companies aren't in a great capital position to add the required electricity production capacity in the time frame they need to. In a rush, many of them are passing along their new build-out costs to consumers in the form of higher rates, which works, but isn't sustainable. In the long run, artificial intelligence data centers need to provide their own power.

Enter GE Vernova (NYSE: GEV).

Yes, this is an offshoot of the old conglomerate you knew as General Electric, which began breaking itself up into more manageable pieces back in 2021. GE Vernova is the power arm of the iconic name, offering everything from wind turbines to grid management solutions to a nuclear power plant service.

Perhaps its most meaningful contribution to the next chapter of the AI revolution, however, is its well-proven, power-generating natural gas turbines.

Although originally built with utility companies in mind, with each one capable of producing a few hundred megawatts' worth of electricity, these sizable turbines are proving to be a practical solution for AI operators looking for self-sufficiency. It's providing 29 of its LM2500XPRESS gas turbines to AI infrastructure outfit Crusoe, for instance, while oil giant Chevron is testing one of its natural gas turbines as a source of power for a utility service intended to specifically serve AI data centers outside of the industry's consumer-facing ecosystem.

For perspective, while this power arm's revenue grew 10% year over year to $5 billion in Q1, it took $10 billion worth of new equipment orders during the same quarter. Indeed, the companywide backlog now stands at $163 billion, compared to Q1's total revenue of only $9.3 billion.

Memory: Micron Technology

There are really only three major computer memory manufacturers -- Micron Technology (NASDAQ: MU), SK Hynix, and Samsung -- and shares of all three are well up since the global memory chip shortage reached critical levels in mid-2025. Be careful buying into any of them.

If you're looking for a specific one to step into on its next opportune pullback, though, Micron is arguably your best bet. Why? Although it's not the biggest in terms of total market share, it is the most focused player in the business and, arguably, best equipped for the future for a couple of reasons.

One of these reasons is how it does business. The company's signing more and more long-term strategic customer agreements, as opposed to more straightforward long-term agreements. As CEO Sanjay Mehrotra commented in March's second-quarter earnings conference call, "We continue to work with customers on strategic customer agreements, or SCAs, that are different from prior LTAs and have specific commitments over a multiyear time horizon for improved visibility and stability in our business model." He added, "We are excited to have signed our first five-year SCA."

The other distinguishing competitive edge is Micron's expertise at designing and manufacturing high-bandwidth memory (HBM) that not only consumes less total power but also occupies less room on a data center circuit board. Micron is sold out of this high-margin memory into 2027, but this dynamic could linger well beyond next year. An outlook from Precedence Research suggests the high-bandwidth memory chip market alone is poised to grow at an average annual pace of more than 25% through 2035.

Optical bandwidth: Marvell Technology

Finally, you probably know that Broadcom is one of the market's most-mentioned data center networking names, largely reflecting its size and long-established business. And if you own (or will own) a stake in this company, you'll be fine.

If you're looking for a name that packs a little more punch because it's smaller and scrappier, though, Marvell Technology (NASDAQ: MRVL) may be your better bet.

But first things first. What's optical bandwidth?

You likely understand that a data center is mostly just thousands of computers linked together into a single, unified "brain." But, in that artificial intelligence data centers are analyzing as well as producing massive amounts of digital data, the routers, switches, and other networking solutions of yesteryear just aren't up to the modern-day task anymore. The industry needs to move information at the speed of light -- literally -- using equipment that sends and receives data using fiber-optic connections. To this end, Goldman Sachs believes the optical networking market will eventually grow ninefold to a $150 billion-plus business.

Marvell makes such equipment. For instance, earlier this month, the company unveiled the industry's first-ever switch capable of processing 102.4 terabits of digital data every second, allowing AI data centers to achieve the next level of computing performance.

Simply being in the right business at the right time doesn't make Marvell a must-have, however. Although it certainly helps, Marvell Technology's distinguishing advantage is that it designs and manufactures complete turn-key AI data center networking systems that include Ethernet controllers, disk drive controllers, and even custom-built computing processors that all seamlessly work together. Last quarter's top-line year-over-year growth of 28% speaks volumes about demand for its technology, as does the 40% growth analysts expect for the full fiscal year.

James Brumley has positions in Alphabet. The Motley Fool has positions in and recommends Alphabet, Amazon, Broadcom, Chevron, GE Vernova, Goldman Sachs Group, Marvell Technology, Micron Technology, and Microsoft. The Motley Fool has a disclosure policy.