Cerebras Is Down 22% Since Its IPO Popped. Should You Buy It Now?

Cerebras has a $24.6 billion backlog and says it can recognize about 15% of that within 24 months, supporting a path to potentially double sales in each of the next two years. The article argues the stock remains expensive at about 108x 2025 sales and highlights execution, customer concentration, and TSMC supply risks despite major deals with OpenAI and Amazon. Shares are already down more than 22% since the May 14 debut after a 68% first-day pop.

Analysis

The market is treating this as a binary story, but the more important read-through is that demand concentration is turning Cerebras into an execution and financing proxy on frontier-model deployment. If the largest buyers are willing to pre-commit multi-year capacity, that validates the architecture, but it also means the equity is now priced like a scaled infrastructure winner before the company has proven it can industrialize delivery, procurement, and uptime at that scale. That combination usually creates a much wider gap between headline bookings and actually realizable revenue than investors model early on.

Second-order beneficiaries are likely the surrounding ecosystem rather than the chipmaker itself. AWS can use the relationship to deepen lock-in with higher-value inference workloads, while TSMC gets incremental utilization on older advanced-node capacity but also inherits a customer with materially weaker allocation power than the mega-cap hyperscalers. The main supply-chain risk is not just wafer supply; it is packaging, data-center buildout, networking, and power delivery, any of which can delay revenue recognition by quarters even if demand remains intact.

The contrarian point is that the pullback may not be enough to reset valuation if the next 12 months are dominated by “proof of installation” rather than “proof of economics.” The stock’s implied multiple still assumes near-perfect conversion of backlog into revenue, yet the business now carries a project-execution profile more akin to a specialized infrastructure roll-out than a pure chip story. That means the next catalyst is not another benchmark win; it is evidence of repeatable deployment cadence and margin durability across tranches.

For NVDA, the competitive threat is more about narrative pressure than immediate share loss: every successful alternative inference stack narrows the premium investors pay for GPU scarcity. But unless Cerebras can demonstrate broad-model economics across multiple customers, the larger implication is that the AI accelerator market is fragmenting into workload-specific winners, which could ultimately support NVDA by keeping total AI spending elevated even as unit economics diversify.

Key Decisions for Investors

• Stay underweight CBRS into the next 1-2 quarters; the setup is still a valuation-vs-execution mismatch, with downside if deployment milestones slip and the market re-rates it toward infrastructure-builder multiples.
• Long AMZN / short CBRS pair for 3-6 months: Amazon can monetize the partnership regardless of whether Cerebras scales flawlessly, while CBRS bears the full operational risk. Favor the pair if hyperscaler capex remains firm.
• Use TSM weakness to accumulate on dips over 1-3 months: Cerebras is not a thesis driver for TSM, but incremental advanced-node demand supports utilization; the risk is capacity prioritization, not structural demand destruction.
• For NVDA holders, hedge a portion of upside with short-dated calls or a small CBRS probe short against the position: the goal is to neutralize narrative risk from alternative-inference headlines without betting against the broader AI spend cycle.
• Avoid chasing CBRS post-rally; if you want upside exposure, wait for either a tranche-delivery update or a 20-30% further pullback, because the current risk/reward still favors disappointment over acceleration.