Rosenblatt maintains Cerebras Systems stock rating on margin beat

Cerebras posted strong results, with revenue up 76% to $510 million in the last twelve months and Q2 2026 non-GAAP gross margin guidance of 37%, well above Rosenblatt’s 24.6% estimate. Full-year gross margin guidance of 39.5% midpoint also topped expectations, while Rosenblatt reiterated a Buy and $300 target. The main constraint appears to be data center availability, but demand remains strong and the company is expanding its cloud footprint globally.

Analysis

The first-order read is not “AI demand is strong” — that is already consensus — but that the bottleneck is shifting from silicon performance to physical deployment capacity. If inference demand remains inelastic and token pricing is preserving premium economics, then the scarcity rent migrates downstream to whoever controls power, racks, and permitted data-center shell capacity. That creates a broader second-order winner set than the headline suggests: utilities, liquid-cooling infrastructure, electrical gear, and colocation providers with available megawatts become the real constraint beneficiaries over the next 6-18 months.

For GOOGL specifically, the signal is mixed but constructive. A healthier inference market can support cloud pricing and improve the relative attractiveness of bundled AI services, but it also validates that compute remains scarce and expensive, which limits near-term margin expansion unless utilization keeps outrunning capex. The more important implication is competitive: if specialized inference pricing is rising, hyperscalers with lower-cost financing and captive power can undercut niche providers over time, compressing the premium as capacity expands. That argues for being selective on which AI beneficiaries own the scarcity rent versus which are paying for it.

The contrarian risk is that the market extrapolates a permanent pricing regime from a temporary supply shortage. If data-center capacity catches up faster than expected, token pricing should mean-revert and the “premium inference” thesis weakens quickly, likely within 2-4 quarters rather than years. In that case, the equity re-rating in smaller AI compute names could stall even if revenue growth stays elevated, because investors will discount the durability of margins rather than the top-line growth rate.

Key Decisions for Investors

• Long GOOGL vs. short a basket of high-multiple AI compute beneficiaries for 3-6 months: own the hyperscaler with balance-sheet optionality while fading names whose valuation assumes persistent scarcity rents; target a 1.5x–2.0x relative-return spread if token pricing normalizes.
• Add a barbell long in data-center enablers (e.g., EQIX, DLR, ETN, VRT) on any post-earnings pullback; thesis is that the real bottleneck is power and deployment, not model demand. Hold 6-18 months; risk is valuation compression if capex spend slows.
• If already long CBRS-type inference names, hedge with short-dated calls or a 1-2 month collar into the next capacity-update cycle; you want to stay exposed to demand but cap downside if investors start discounting margin mean reversion.
• Watch for confirmation in utility and grid names over the next 1-2 quarters; a sustained breakout would validate the physical-capacity bottleneck and offer a cleaner way to express the AI buildout than betting on any single chip vendor.