Artificial intelligence has looked like a software story for the past few years. Investors bought GPUs, cloud platforms, and model companies that can run AI models for everyday users. But there is another part of the AI boom that is becoming harder to ignore: AI requires a huge amount of physical infrastructure.
A server cannot run a large AI model without power. A rack packed with accelerators cannot operate without moving enormous amounts of data or removing the heat those chips generate. That creates an interesting opportunity for investors who want exposure to AI without buying the same handful of megacap stocks.
Three companies stand out to me. They operate in very different markets, but all three are attacking physical bottlenecks created by increasingly powerful AI systems.
Image source: Getty Images.
1. Astera Labs is solving the data movement problem
Astera Labs (ALAB +9.75%) may be one of the more overlooked companies in the AI infrastructure chain. Its technology sits between computing components, helping move data between CPUs, GPUs, memory, and networking equipment. That sounds less exciting than designing a GPU, but it becomes critical as AI systems grow more complex.
Astera’s newest Scorpio X-Series 320-lane fabric switch is designed for rack-scale AI systems and is built to connect large numbers of accelerators. The company has also expanded into CXL memory controllers, Ethernet connectivity, and optical interconnects. In other words, Astera is trying to become a broader connectivity platform rather than relying on a single product.
That is the part I find compelling. AI performance increasingly depends on keeping processors fed with data. Faster chips do not help much when information cannot move between them efficiently.
Today’s Change
(9.75%) $27.58
Current Price
$310.40
Key Data Points
Market Cap
Day’s Range
$285.84 – $321.65
52wk Range
$97.89 – $499.48
Volume
6.5M
Avg Vol
5M
Gross Margin
75.13%
2. Modine is betting that AI has to be cooled
Modine Manufacturing (MOD +4.30%) is an even less obvious AI stock. Its opportunity is simple: Powerful AI servers generate a lot of heat.
Modine’s Airedale business makes cooling systems designed for data centers, including its new TurboChill 3+MW platform for high-density GPU environments. The company also entered into a long-term agreement worth more than $4 billion for Airedale cooling products through 2029, backed by a $165 million upfront payment from the customer to help fund capacity investments.
That customer commitment is important because cooling is not an optional feature that hyperscalers can remove when budgets get tight. As computing density increases, thermal management becomes part of the facility’s basic architecture. Modine even created a dedicated data center segment this year, a sign that management sees the opportunity as large enough to deserve its own focus.
Today’s Change
(4.30%) $8.03
Current Price
$194.66
Key Data Points
Market Cap
Day’s Range
$186.10 – $195.00
52wk Range
$111.18 – $323.25
Volume
1.3M
Avg Vol
1.5M
Gross Margin
22.38%
3. Flex is helping turn AI designs into real machines
Flex (FLEX +1.50%) may be the most practical name on this list.
The company manufactures complex electronic systems, and it is now applying that manufacturing expertise to AI infrastructure. Its AI factory reference designs combine power, high-density racks, networking, and liquid cooling into modular systems designed to accelerate deployment. Flex says the approach can reduce deployment timelines by up to 30%.
It is also manufacturing Advanced Micro Devices’ instinct AI systems in the United States, including complete platforms with eight AMD GPUs, high bandwidth memory, high-speed interconnects, and liquid-cooling hardware supplied by its JetCool business. That gives Flex an interesting position. It is not trying to invent the next AI model. It is helping turn someone else’s AI architecture into a physical system that can be delivered to a customer.
Today’s Change
(1.50%) $1.62
Current Price
$109.51
Key Data Points
Market Cap
Day’s Range
$107.00 – $111.63
52wk Range
$53.07 – $166.86
Volume
4.5M
Avg Vol
5.8M
Gross Margin
9.29%
How I’d invest $1,000
I would not treat these as three interchangeable AI bets. I would put roughly $350 into Astera Labs, $350 into Modine, and $300 into Flex.
Astera offers the highest exposure to the semiconductor connectivity layer. Modine gives investors a bet on one of AI’s unavoidable physical constraints: heat. Flex provides exposure to the manufacturing and systems integration aspects of the build-out. The AI era is becoming physical. Billions of dollars are being spent on servers, power systems, cooling equipment, networking, and factories. Investors do not have to own the company designing the smartest AI model to benefit from that spending.
