Beyond NVIDIA: 2 AI Data Center Stocks to Buy for More Upside

AI data-center infrastructure demand is broadening beyond NVIDIA’s GPUs, supporting power, cooling and networking suppliers such as Vertiv and Arista. Vertiv reported 24% year-over-year Q2 2026 sales growth, guided Q3 sales to $3.65-$3.85 billion, and agreed to acquire UtilityInnovation Group for $1.45 billion upfront plus up to $1.15 billion contingent consideration; analysts project 2026 EPS and revenue growth of 59.3% and 36.9%, respectively. Arista’s Q2 revenue rose 37.7% to more than $3 billion and it guided Q3 revenue to about $3.3 billion, with 2026 earnings and revenue projected to grow 35.6% and 39.1%.

Analysis

The relevant differentiation is not simply "AI infrastructure" but the bottleneck each vendor monetizes. VRT is levered to physical deployment constraints—power availability, heat rejection and time-to-energize—where project complexity can raise switching costs and support service revenue. ANET monetizes cluster scale-out, but faces a more direct technology and pricing contest from AVGO’s switching silicon ecosystem, NVIDIA’s InfiniBand/Ethernet roadmap and white-box alternatives; its premium multiple therefore requires sustained share gains, not merely industry growth.

The UtilityInnovation acquisition increases VRT's strategic value if data-center operators move from grid-connected builds toward behind-the-meter generation and microgrids. It also introduces execution risk: the contingent consideration implies aggressive EBITDA targets, while a cash-funded acquisition ahead of closing can pressure leverage and obscure organic margin conversion. The more investable second-order beneficiaries of power scarcity may be Eaton (ETN), GE Vernova (GEV), Quanta Services (PWR), Cummins (CMI) and Caterpillar (CAT), whose electrical equipment, transmission and onsite-power exposure can benefit before a data center reaches VRT's cooling-install phase.

Over the next 1-3 months, hyperscaler capex commentary and lead-time data are the key catalysts; an unchanged capex budget is insufficient if project energization delays push equipment acceptance into later quarters. Over 6-18 months, the risk is a capex digestion cycle: GPU orders can be rephased quickly, while infrastructure suppliers retain working-capital and fixed-cost exposure. The contrarian view is that the networking leg is more consensus-owned than the power-constraint leg; favor VRT only if order growth and backlog conversion remain intact, rather than treating its drawdown as evidence of valuation support.

Thesis fails if hyperscaler capex guidance is cut, VRT's book-to-bill falls below 1.0x or its gross-margin outlook declines despite revenue growth; for ANET, watch customer concentration, AI-networking attach rates and any evidence that NVIDIA/AVGO solutions compress Ethernet pricing. A broad risk-off move in high-duration AI equities can overwhelm these company-specific signals in the near term.

Key Decisions for Investors

• Accumulate VRT in tranches over the next 1-3 months only after confirming stable backlog/book-to-bill and no reduction in FY margin guidance; target a 12-18 month re-rating on power-constrained deployments. Size modestly until acquisition financing and expected leverage are disclosed; exit on sub-1.0x book-to-bill or a material gross-margin guide-down.
• Run a 3-6 month pair: long VRT / short ANET on equal dollar beta-adjusted exposure. This expresses the view that physical power and cooling scarcity has more durable pricing power than increasingly competitive AI fabrics; reassess after each company's next earnings call and cover if ANET demonstrates accelerating AI revenue with stable gross margin.
• For a broader infrastructure basket, overweight ETN and PWR versus HPE for 6-18 months. ETN/PWR capture electrical distribution and interconnection spend irrespective of server vendor share, while HPE remains exposed to lower-margin systems integration and enterprise procurement cycles; invalidate if utility interconnection timelines meaningfully improve or data-center construction starts roll over.
• Do not add directional ANET exposure after momentum alone. Set an alert for evidence of Ethernet AI-cluster share gains versus NVIDIA/AVGO and for sustained gross-margin resilience; absent that data, use a break above post-earnings highs with positive revisions as the entry trigger rather than chasing the current AI-networking narrative.