Short Seller Culper Research Targets Chipmaker Cerebras
nashnova research
Short-seller Culper Research alleges Cerebras's core wafer lifespan is just 1.5–2.5 years, far below the company's claimed 3–5 years; if true, the commercial foundation behind its $20 billion valuation is at stake.
What exactly is Culper Research alleging?
Culper Research published a short report on October 28, titled "Cerebras Systems: Hot Garbage."
The core claim is narrow: Cerebras's Wafer-Scale Engine (WSE) — a single full silicon wafer turned into one enormous chip — has an actual lifespan of just 1.5–2.5 years, not the 3–5 years the company advertises.
This means → if the wafer lasts half as long, customers must replace it twice as often, doubling their effective cost and undermining Cerebras's value proposition.
How did Cerebras respond?
Cerebras fired back on its official blog, saying "nearly all" of Culper's conclusions rest on the assumption that wafer life is short — "but wafer life is not short; it is long."
The company says it has "nearly perfectly solved thermal expansion" through years of process engineering that dampens the impact of temperature swings on the wafer.
In plain terms = both sides are arguing over how long the same chip lasts; Cerebras says it spent years specifically solving that problem, and the short-seller's premise is simply wrong.
Why was the CEO's track record brought up?
Culper's report also questioned co-founder and CEO Andrew Feldman's history.
In the early 2000s Feldman was VP of marketing at Riverstone Networks, which later filed for bankruptcy. The SEC charged him with involvement in misstating net income; he pleaded guilty to one felony count of circumventing internal accounting controls.
He later co-founded low-power server startup SeaMicro, sold it to AMD for $334 million in 2012, but the product line was shut down three years later under CEO Lisa Su, and AMD took a $75 million write-down.
This reflects a two-front strategy by Culper — product-lifespan doubt on one side, management-credibility doubt on the other, pressing market confidence from both directions.
How far did the stock fall — and what does the market think?
Cerebras currently carries a market cap of roughly $20 billion. Its shares surged 70% on IPO day, then retreated from a $350 high to around $81.71.
By the morning after the short report, the stock had fallen about 15% — yet the iShares Semiconductor ETF dropped roughly 17% over the same window, meaning Cerebras slightly outperformed the sector.
This means → the market did not fully side with the short-seller. Part of the reason: Cerebras had just posted a strong Q3 report and signed new partnerships with Microsoft, Oracle, and Amazon, helping reduce its revenue concentration on OpenAI.
What should investors watch next?
The central dispute — wafer life of 3–5 years or 1.5–2.5 years — remains a he-said-she-said standoff that cannot be directly verified from public data.
In plain terms = upcoming earnings and customer-renewal data are the decisive window: if major clients renew on a 3–5 year cycle, the short thesis loses its footing; if they don't, Cerebras faces real valuation pressure.
市场有风险,内容仅供研究参考,不构成投资建议。
