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

Cerebras shares are down more than 22% since their May 14 debut despite a 68% first-day pop, as the article argues the stock still looks expensive at about 108x 2025 sales and 27x 2027 sales even under a doubling-sales scenario. The bullish case rests on a $24.6 billion backlog, including a $20 billion OpenAI deal and Amazon AWS partnership, but execution, customer concentration, and TSMC capacity risk remain significant. The piece concludes the stock is not attractive at current levels despite strong technology and large potential revenue visibility.

Analysis

The market is still pricing Cerebras like a pure-IPO scarcity story, but the more important implication is that its design is moving from novelty to an ecosystem enabler. If its latency advantage holds in chained inference and agentic workflows, the first-order winner is not just Cerebras but the cloud platforms that can monetize lower-cost reasoning sessions; that is why AMZN looks better positioned than a standalone chip beneficiary, while NVDA’s moat is more exposed at the high-end inference layer than the stock suggests.

The second-order supply-chain issue is TSMC. Cerebras does not just compete for wafers; it competes for process attention and yield support on a legacy node that is increasingly crowded by larger, higher-priority customers. That makes the key constraint less “can demand exist?” and more “can capacity be delivered on schedule?”—a classic execution bottleneck that can compress expected revenue recognition even if the end demand is real.

Consensus seems to be underweighting how concentrated contract value can become a fragility rather than a strength. A single large customer can validate the platform, but it also creates path dependence: any slip in deployment cadence can trigger a re-rating because the equity’s multiple is carrying multiple years of flawless conversion, not just signed backlog. That asymmetry makes the stock vulnerable to a very sharp drawdown on any missed tranche or manufacturing delay, while upside requires a long sequence of clean operational milestones.

From a trade perspective, the cleanest expression is not to short the technology thesis, but to fade the valuation-vs-execution mismatch. The better risk/reward is to own AMZN or NVDA rather than chase a binary, high-multiple IPO; if Cerebras works, both can participate in broader AI spend, but if it stalls, they are insulated. TSM is the quiet barometer: any capacity strain there raises the probability of delay across the whole story and could pressure suppliers first before the chipmaker itself fully re-prices.

Key Decisions for Investors

• Avoid initiating a fresh long in CBRS at current levels; wait for either a post-lockup/earnings reset or evidence of on-time tranche delivery. Near-term upside is capped by execution risk, while downside can re-rate quickly on any delay.
• Add to AMZN on pullbacks over the next 1-3 months: it has the best asymmetric exposure to inference disaggregation and can monetize Cerebras-like workloads without taking single-name manufacturing risk.
• Hold or selectively add NVDA, but use CBRS weakness as a read-through that the inference market is bigger than one architecture, not as a reason to rotate out. NVDA remains the cleaner long because its revenue visibility is far better.
• Use TSM as a hedge indicator: if there is chatter about 5nm capacity/yield tightness, consider a tactical long TSM vs short a basket of high-beta AI infrastructure names for 4-8 weeks, as capacity bottlenecks would hurt the most execution-sensitive names first.
• If trading CBRS at all, prefer a small-size put spread 3-6 months out rather than outright short stock. The setup is a classic high-multiple/operational-risk name where downside can accelerate, but borrow and squeeze risk make directionally shorting unattractive.