5 Under the Radar AI Chip Stocks Powering the Data Center Boom

The article argues the AI datacenter buildout is shifting beyond GPUs into RF/optical interconnects, custom silicon, licensed CPU cores, PCIe fabrics, and wafer-scale inference—highlighting sustained revenue acceleration across the “plumbing” stack. Examples include MACOM’s Q2 FY2026 revenue of $288.95M (+22.5% YoY) and 58.5% adjusted gross margin, Marvell’s Q1 FY2027 revenue of $2.418B (+27.6% YoY) with data center at $1.83B (+27%), Arm’s Q4 FY2026 revenue of $1.49B (+20.1% YoY) with licensing revenue up 29%, Astera’s Q1 2026 revenue of $308.36M (+93.4% YoY) and EPS $0.61 vs $0.54 expected, and Cerebras’ IPO/contract backdrop including a $20B+ OpenAI inference deal and $6.4B IPO proceeds. While valuations appear stretched (e.g., Arm at 137x forward earnings) and Cerebras shares are down ~30.9% from the IPO peak, the overall tone is that the group’s demand and guidance are strengthening and sell-side targets are catching up.

Analysis

The market is still pricing AI as a GPU-only story, but the higher-probability upside is in the bill of materials around the compute node. That matters because connectivity and control silicon usually scale with rack density and port speed even when headline accelerator unit growth moderates, so names like MTSI, ALAB, and MRVL can compound off a broader installed base with less direct exposure to the next model-cycle wobble. ARM remains the cleanest royalty toll-booth, but it is also the most crowded expression of the theme and the most vulnerable to multiple compression if expected design wins slip.

The near-term catalyst path is earnings and backlog over the next 1-2 quarters: any evidence that 1.6T optics, CPO, PCIe/CXL, or custom CPU ramps are translating into sustained guide raises should keep the group bid. The main reversal risk is a capex pause driven by power constraints, procurement delays, or ROI scrutiny; that would hit the smaller, more execution-sensitive names first and likely compress the highest-beta valuations before it shows up in end-demand. Over 6-18 months, the winners should be the vendors with the highest attach-rate per rack and the cleanest path to recurring revenue, not necessarily the fastest top-line growers today.

Contrarian view: the consensus is treating all AI plumbing as additive, but some of it is substitution within the stack rather than net-new spend. That means the cheapest apparent growth may actually deserve the highest multiple if it is attached to durable design wins, while the highest-growth names can still underperform if gross margin, customer concentration, or financing terms deteriorate. CBRS is the clearest example: the equity story depends on a small number of hyperscale relationships and the market will punish any slippage in deployment timing far faster than it rewards the headline contract size.

Key Decisions for Investors

• Long MTSI on pullbacks over the next 2-6 weeks; it is the least crowded way to express rising optical content per rack, with a better risk/reward than the more saturated AI leaders. Falsifier: sequential growth or margin guide stalls on the next print.
• Pair trade: long MRVL / short ARM for the next 1-3 months. MRVL has more directly verifiable data-center revenue and lower valuation risk, while ARM is more exposed to multiple compression if royalty growth normalizes. Target 15-20% relative outperformance if AI capex broadens but sentiment cools.
• Long ALAB versus short CBRS as a quality-vs-speculation relative-value trade. ALAB monetizes standardized rack-level complexity; CBRS depends more on execution, financing optics, and a narrow set of deployment outcomes. Keep size modest; cover the short if CBRS shows repeatable margin improvement and customer milestones.
• Do not chase CBRS outright after the post-IPO pullback; treat it as a watch item until the next update confirms shipment cadence and cash needs. The risk/reward only improves if the market gives you a larger discount without fundamental deterioration.