Cerebras Systems vs. Nebius Group N.V.: Which AI Hardware Stock Is a Better Buy in 2026?

Nebius is presented as the preferred high-risk AI infrastructure investment for 2026, supported by lower valuation multiples than Cerebras at 49.0x forward P/E and 47.6x sales, versus Cerebras at 222x and 71.2x. Nebius generated roughly $530 million of FY2025 revenue, up 351% year over year, but burned about $3.7 billion in free cash flow; Cerebras produced $510 million of revenue, up 76%, with $238 million of net income but negative $393 million free cash flow. Analyst projections are highly speculative, implying Nebius revenue could rise to $23 billion by 2028 and Cerebras revenue to $7.5 billion, while both face intense competition from hyperscalers and established AI-chip providers.

Analysis

The central underwriting issue is not relative valuation but capital intensity per dollar of durable revenue. NBIS requires sustained data-center, GPU/networking, and power investment before customer contracts convert into cash returns; a high reported growth rate can therefore coexist with worsening equity dilution or leverage. Its lower forward P/E is not a margin-of-safety signal if consensus is capitalizing earnings before depreciation, financing costs, and replacement capex normalize. CBRS has the inverse profile: specialized hardware can generate attractive incremental gross margins, but its revenue base is more exposed to customer concentration, design-win timing, and an architecture-specific obsolescence event.

Over the next 1-3 months, both names will trade more on AI infrastructure risk appetite and disclosed backlog/contract duration than on trailing profitability. The key differentiator is whether NBIS can demonstrate that capacity is pre-sold at returns above its cost of capital, while CBRS must show that inference demand is recurring rather than benchmark-driven pilot activity. NVDA remains the second-order beneficiary in either outcome: a cloud buildout by NBIS increases accelerator and networking demand, while a credible CBRS share gain would be a narrow but sentiment-relevant challenge to NVDA's inference multiple.

Consensus appears to be treating multi-year sales forecasts as execution paths rather than requiring a financing bridge. The more likely downside scenario is not demand disappearance but lower pricing and utilization as hyperscalers use excess capacity and bundled cloud credits to defend enterprise workloads; that would impair NBIS first. CBRS's elevated sales multiple leaves little room for a missed ramp, but its smaller physical footprint could make it the better asymmetric vehicle if independent workload benchmarks establish materially better cost-per-token economics.

For 6-18 months, monitor capex-to-revenue, committed versus deployed capacity, customer concentration, gross-margin trajectory, and stock-based-compensation dilution. Thesis failure for NBIS is two consecutive quarters of utilization or gross-margin deterioration alongside rising net debt; for CBRS, it is a delayed major deployment, weaker cloud-service mix, or third-party benchmarks showing no sustained inference cost advantage versus NVDA-based systems.

Key Decisions for Investors

• Avoid outright NBIS length until the next earnings release discloses contracted capacity, utilization, and financing needs; initiate only if incremental capacity is demonstrably pre-sold and gross margin holds or expands. A 20%+ drawdown is plausible if capex guidance rises without corresponding backlog conversion.
• Prefer a small, catalyst-driven long CBRS / short NBIS pair over 3-6 months, sized dollar-neutral: CBRS has greater upside to independently verified inference performance, while NBIS is more vulnerable to capital-market repricing. Exit if CBRS misses a material deployment milestone or NBIS reports improving utilization with funding fully secured.
• Maintain NVDA as the higher-quality AI infrastructure exposure rather than replacing it with either name. Use any CBRS benchmark win as an alert to trim, not structurally short, NVDA unless customer deployments show repeatable share loss.
• For existing NBIS holders, cap exposure ahead of earnings and treat a rise in debt, share count, or capex without a matching increase in contracted revenue as a sell trigger; the valuation framework should be EV to normalized free cash flow, not headline forward P/E.