3 AI Stocks Poised to Outperform Micron and Sandisk as the Next Infrastructure Bottleneck Builds

The article highlights Innodata, Nebius and Vertiv as potential AI-infrastructure beneficiaries beyond memory chips. Innodata posted 58% year-over-year Q2 revenue growth and guided to 40% full-year growth, while Nebius's cloud revenue rose 454% to $582.3 million and annual contract value per megawatt reached $20 million, with some prospective deals above $40 million. Vertiv reported 24% Q2 revenue growth and expects 31% full-year growth as liquid-cooling demand rises with increasingly power-intensive AI deployments.

Analysis

NBIS is the highest-beta expression of the AI power bottleneck, but the valuation debate should center on financed, grid-connected capacity rather than its development pipeline. Premium short-duration capacity pricing can drive upside over the next 1-3 quarters, yet it is not equivalent to durable ARR: renewals will be exposed to GPU utilization, hyperscaler self-build, power curtailment provisions, and the cost of debt required to convert land/power options into operating sites. A widening between contracted capacity and energized capacity, or lower customer prepayments, would be the earliest signs that the narrative is outrunning cash conversion.

VRT has the cleaner 6-18 month earnings setup because higher rack density raises cooling content per megawatt and shifts customer buying toward integrated thermal/power systems, where qualification cycles protect margins. The less obvious beneficiary is electrical infrastructure—ETN, PWR and HUBB—because liquid cooling does not eliminate the need for switchgear, distribution, interconnection and backup power. The key risk is that VRT's premium multiple embeds sustained order conversion; any backlog-to-revenue slippage or gross-margin erosion from project mix would compress the stock quickly.

INOD's improved customer mix reduces a prior concentration discount, but frontier-lab contracts can be volatile and labor-intensive. The equity needs evidence that new accounts produce recurring, higher-value data curation work rather than episodic model-training projects; otherwise revenue growth may not translate into operating leverage. Consensus may be underappreciating that data-services spend is a smaller and more discretionary budget line than power or cooling when AI capex is rationalized.

Key Decisions for Investors

• Maintain an overweight in VRT for the next 2-4 quarters; add on post-earnings weakness only if orders, book-to-bill and gross-margin guidance remain intact. Falsify on two consecutive quarters of backlog growth deceleration combined with margin-guide cuts; prefer VRT over NVDA as a second-order AI-capex exposure.
• Use a defined-risk NBIS call spread, 6-9 months out, rather than common equity: buy only after verification of incremental energized capacity, financing terms and customer prepayments. Target asymmetric upside from additional premium-capacity contracts; exit if net debt/capex needs rise materially faster than contracted revenue.
• Pair long VRT against short a broad AI-compute proxy such as SMH only if cooling-order growth continues to exceed semiconductor-equipment order growth. The trade monetizes the shift from chip scarcity toward physical deployment constraints; stop out if hyperscaler capex guidance is cut or VRT order intake misses.
• Keep INOD on a watchlist rather than initiate aggressively. Upgrade to a long only after disclosure shows new-customer revenue scaling and sustained operating-margin expansion; a renewed increase in top-customer concentration or a guide below roughly 40% growth would invalidate the re-rating thesis.
• Monitor ETN, PWR and HUBB as lower-volatility beneficiaries of data-center power buildouts. A sustained rise in grid interconnection delays strengthens their pricing power but weakens NBIS's near-term commissioning timeline.