Cerebras Systems vs. Innodata: Which AI Infrastructure Stock Is a Better Buy in 2026?

Cerebras reported FY2025 revenue of nearly $510.0 million, up 75.7%, and net income of $237.8 million, but generated negative free cash flow of $392.8 million and trades at 137.9x forward P/E and 59.7x sales. Innodata generated FY2025 revenue of $251.7 million, up 47.6%, with $32.2 million of net income and $35.6 million of free cash flow, while trading at lower multiples of 58.3x forward P/E and 7.4x sales. The article favors Innodata for its consistent growth, improving profitability and lower valuation, while highlighting customer concentration and litigation risks; Cerebras faces spending, scaling and competitive hardware risks despite its OpenAI and AWS partnerships.

Analysis

The relevant distinction is not hardware versus services, but whether AI spend is capex-funded and capacity-constrained (CBRS) or operating-expense-funded and labor-sensitive (INOD). CBRS's reported accounting profitability is not yet economically validated by cash conversion: persistent large negative FCF alongside a balance-sheet deficit means incremental cloud/customer deployments may require external capital before they become self-funding. At an extreme revenue multiple, even modest delays in utilization, receivables collection, or wafer supply can drive sharp multiple compression over the next 1-3 quarters.

INOD has a more defensible near-term earnings setup only if customer diversification is observable in reported bookings rather than management commentary. Its largest-account exposure creates asymmetric downside: a procurement insourcing decision by a hyperscaler or frontier lab could remove enough volume to overwhelm margin gains from new logos. Conversely, an expanding mix of recurring model-evaluation work would improve revenue durability and support a rerating, but labor-intensive delivery means wage inflation and SBC dilution will cap operating leverage over 6-18 months.

Consensus is likely underpricing the possibility that frontier-model builders vertically integrate data curation while simultaneously diversifying compute beyond GPU clusters. That creates a difficult middle ground for both names: CBRS must prove workload-specific cost/performance and high fleet utilization, while INOD must prove it owns proprietary workflow/IP rather than simply providing scalable labor. NVDA remains the cleaner AI-infrastructure exposure until CBRS discloses contract economics, backlog conversion, and capex funding requirements.

Key Decisions for Investors

• Do not chase CBRS on growth headlines. Use a 1-3 month earnings/filing watch: consider a tactical short only if operating cash burn remains above 50% of revenue or receivables rise materially faster than sales; cover if contracted backlog converts into positive operating cash flow or disclosed cloud utilization materially improves.
• Initiate INOD only as a sized, catalyst-driven long after the next quarterly customer-concentration disclosure confirms the largest customer is below 50% of revenue and guidance is maintained or raised. Target 6-12 months; invalidate on a major-customer volume decline, flat sequential AI-services revenue, or SBC exceeding operating cash flow growth.
• For relative-value exposure, prefer long NVDA / short CBRS in equal volatility-adjusted notional over the next 1-3 quarters. The trade monetizes CBRS execution and financing risk while retaining AI-compute demand exposure; stop if CBRS demonstrates sustained positive FCF and a clearly superior cost-per-token profile at commercial scale.
• Avoid treating INOD's reported FCF as fully distributable. Model cash generation after SBC and wage normalization; if adjusted FCF yield remains unattractive versus ACN, INOD's multiple rerating case depends entirely on growth persistence rather than cash-return capacity.