The global artificial intelligence (AI) infrastructure isn't built only with cutting-edge chips or advanced software. It also depends on fast and low-latency networking infrastructure to connect thousands of components inside modern data centers. In fact, the data center networking market is expected to grow from $39.7 billion in 2025 to $118.9 billion in 2033.
A few smaller companies are quietly capturing share in this market. Here's why these three stocks below could pleasantly surprise investors in the long run.
Image source: Getty Images
Astera Labs
Astera Labs (ALAB 1.91%) is focusing on the cloud and AI connectivity bottlenecks in the fast-growing AI infrastructure market. The company's open-standard PCI Express (PCIe) solutions are being widely adopted in data centers to enable high-speed and low-latency communication between CPUs, GPUs, memory, and storage within AI servers and across AI servers in a rack.
The company's portfolio includes Aries retimers (cleaning and amplifying signals), Scorpio switches (connecting PCIe components within a server or data rack), Taurus electrical cables, and Leo CXL controllers (for a new standard built on PCIe). All of them together reduce noise and signal loss in data communications.
Astera Labs, together with Parade Technologies and Analogix, accounts for nearly 86% share of the global PCIe retimer market. With the global retimer market expected to grow from $613.6 million in 2024 to over $1 billion in 2029, Astera Labs may see significant upside in the coming years.
The company estimates that its addressable market will expand from $17.2 billion in 2024 to $27.4 billion in 2027. The company is also working closely with partners such as Nvidia, Advanced Micro Devices, and Alchip to co-develop and validate connectivity solutions. This has positioned it as a significant beneficiary of the industrywide shift toward open, multi-vendor, and scalable data center architectures.
Astera's solutions are also becoming mission-critical, as hyperscalers upgrade network infrastructure from 400-gigabits (400G) to 800-gigabits (800G) and even 1.6-terabits (1.6T). Since large amounts of data must be moved around to train and run complex AI models, data centers are increasingly opting for faster, low-latency, and higher-throughput networking solutions
Revenue and earnings are growing at a rapid pace. Astera also has over $1 billion in cash on its balance sheet at the end of the second quarter (ended June 30). But the stock is definitely not cheap, trading at nearly 102 times forward earnings. However, such valuations are not rare for early-stage growth companies riding robust tailwinds.
Hence, Astera Labs can emerge as an impressive AI infrastructure pick for long-term investors, even at current elevated valuation levels.
Celestica
Celestica (CLS -0.19%) is a lesser-known name in the AI infrastructure market, but it is definitely not less important. The company designs, engineers, and builds critical components required for an AI infrastructure buildout, including networking switches, server platforms, and full-rack systems.
One of the most significant catalysts for Celestica is the upgrade of AI data center clusters from 400G Ethernet interconnects to 800G and 1.6T. Broadcom's Tomahawk 5 switches are used extensively in this transition. As a longtime manufacturing and supply partner of Broadcom, Celestica may benefit from the latter's recent $10 billion custom chip supply deal with OpenAI.
Demand has caused revenue for Celestica's Connectivity & Cloud Solutions (CCS) segment to soar 28% year over year to $2.07 billion in the second quarter of fiscal 2025 (ended June 30). CCS now accounts for 72% of the company's revenue.
The higher margin Hardware Platform Solutions (HPS) networking business within CCS also saw revenue surge 82% year over year to $1.2 billion. The company also generated $120 million in free cash flow and ended the second quarter with over $1 billion in liquidity. Celestica thus has sufficient funds for future capacity expansion.
Currently, the shares trade at 36.9 times forward earnings, which may be concerning to some investors. However, considering the robust AI-driven demand and improving revenue mix, the future seems bright even at higher valuations.
Fabrinet
Fabrinet (FN -0.34%) is another under-the-radar company that is helping power the AI infrastructure buildout. The company provides optical communication devices, including selective switching products, high-speed transceivers (which convert electric signals into light and back), interconnects, and optical cables, as well as advanced optical packaging. The laser-based communication hardware facilitates faster and lower-latency server communication, which is essential for running complex AI workloads.
Data centers are upgrading from 800G to 1.6T optical links. Maintaining precise thermal control and alignment of optical components inside transceivers is crucial to prevent signal loss, errors, or overheating. This, however, can prove challenging when data is transferred at high speeds such as 800G or 1.6T.
Since Fabrinet is one of the few contract manufacturers that has consistently demonstrated this capability, it stands to benefit from this network upgrade trend. The company has started shipping volumes of 1.6T transceivers, and expects demand for 800G and 1.6T transceivers to remain strong in fiscal 2026.
With demand significantly outpacing supply, Fabrinet is facing short-term supply shortages for critical components. However, management anticipates that this will be a temporary challenge. These supply issues are to be resolved within the next one to two quarters. In fiscal 2025 (ended June 27), Fabrinet's revenue rose 19% year-over-year to a record $3.4 billion, while non-GAAP (adjusted) earnings per share also reached a record high of $10.17.
The company also has several reputable customers, including Nvidia and Cisco Systems. But it trades at almost 40 times forward earnings, which is expensive. Yet, with its competitive moat and rapidly expanding market opportunity, the stock appears to be a good pick for long-term investors.
