Cerebras lock-up ends before $214M-$216M Q3 results

By Nathan Brooks

Cerebras Systems’ main IPO lock-up expired at 6:00 a.m. ET on November 9, three days before its third-quarter 2026 results. CBRS traded at $93.79 at 11:44 a.m. ET, up 4.58%, but remained 49.3% below its $185 IPO price.

CBRS had reached an intraday low of $77.84 on November 5 before closing at $89.68 on November 6. Earnings arrive after the Thursday, November 12 market close, with a conference call at 5 p.m. ET.

Why does the lock-up end on November 9?

Cerebras’ final IPO prospectus, filed with the SEC on May 14 and dated May 13, 2026, sets the main lock-up at the earlier of two dates: 6:00 a.m. ET on the second trading day after results for the quarter ended September 30, or 180 days after the prospectus date.

The 180-day calculation produces November 9. The earnings alternative would be November 16: Friday, November 13 is the first trading day after the November 12 announcement, and Monday, November 16 is the second. The earlier date controls.

MarketBeat identified November 10, apparently counting from the May 14 filing rather than the May 13 prospectus date. The one-day discrepancy has a paper trail.

How much stock can actually trade?

Lock-up agreements covered Cerebras, its executive officers and directors, and certain other holders representing about 71.4% of its Class A common stock and related convertible securities. Other holders faced separate market-standoff provisions.

Cerebras estimated that up to approximately 171.1 million shares could become eligible through early-release provisions across the full lock-up period. The aggregate includes up to 15 million shares held by directors and officers; it is not a November 9-only tally.

The final provision may make remaining, previously ineligible stockholder shares available, subject to Rule 144 and affiliate limits. The contract has ended, but permission is not an order to sell.

What did insiders already sell?

CTO Sean Lie sold 710,247 shares on August 20 for gross proceeds of approximately $147.63 million under a Rule 10b5-1 plan adopted May 20. The completed sale was not a new proposal tied to November 9.

A Form 144 proposed a separate sale of 60,000 Lie shares with an estimated market value of about $6.5 million on October 27; his next-day Form 4 confirmed all were sold that day under the same prearranged plan.

COO Dhiraj Mallick sold 396,000 shares on September 29 for about $80.8 million under a Rule 10b5-1 plan adopted June 30. His same-day Form 144 proposed the sale; a September 30 Form 4 confirmed execution.

An August 18 Form 4 separately recorded Mallick selling 38,889 shares for approximately $8.64 million. It classified the sale as non-discretionary sell-to-cover for tax withholding after restricted stock units settled.

The distinction matters: a completed trade is history, a Form 144 may describe a proposal, and tax withholding is not a discretionary view on the stock. Both executives’ filings also covered other transactions tied to equity compensation. None proves an immediate post-lock-up sale plan.

What must the November 12 results show?

Management guides to third-quarter core revenue of approximately $214 million to $216 million, core gross margin of 38% to 40%, and core operating margin of negative 25% to negative 23%. These are non-GAAP core measures, not GAAP results.

In the second quarter, Cerebras posted GAAP revenue of $180.1 million and core revenue of about $209.9 million. Core gross margin was 41%, against roughly 14% GAAP.

The gap includes customer warrant amortization, stock-based compensation, data-center pass-through revenue and costs, and other specified adjustments. Investors need to see whether revenue growth and rapid AI-inference expansion produce durable improvements in underlying economics as Cerebras expands manufacturing and data-center capacity.

Customer commitments sharpen the test. At June 30, remaining performance obligations stood at $25.4 billion, with a significant portion tied to OpenAI.

OpenAI committed to buy 750 megawatts of AI inference capacity and holds an option for another 1.25 gigawatts. Cerebras recognized $56.8 million from the arrangement in the second quarter, net of customer warrant amortization.

The report should show how much contracted capacity is operating, how quickly it deploys and how much backlog becomes recognized revenue. Cerebras also said more than 600 megawatts of data-center capacity would be live or under contract for delivery by the end of 2027. Turning commitments into usable capacity without excessive cost increases matters because the build requires substantial investment.

Can Cerebras settle the wafer-life dispute?

On October 28, Culper Research disclosed a short position and argued that Cerebras’ wafer-scale processors have a useful field life of roughly 1.5 to 2.5 years. It questioned Cerebras’ proprietary connector claims, alleging use of older third-party material called PariPoser, and said continuous dummy operations prevent damaging temperature changes while raising power use.

Cerebras answered on November 5 in a statement titled “What It Took to Build Wafer-Scale, and Why It Lasts.” CEO Andrew Feldman rejected the two-year estimate and said “the company had never shipped a system using PariPoser.”

He described controlled power consumption and temperature management as deliberate engineering designed to limit mechanical stress across the wafer. Cerebras said less than 0.5% of deployed CS-3 wafers had been decommissioned since the system entered service in 2024.

Feldman said TSMC qualifies the wafer for 10 years of continuous operation at 105°C, while the deployed CS-3 fleet’s average junction temperature remains below 65°C. These are company claims, not independent verification.

Cerebras changed its accounting estimate for internally used equipment from one-to-three years to three-to-five as assets shifted from research and development to production systems supporting the cloud business. Feldman tied the revision to the operating environment and said it remains subject to annual audits.

The accounting comparison is direct: three-to-five years on the books against an alleged 1.5-to-2.5-year field life. Shorter service lives could impair hardware economics, customer returns and financial assumptions supporting long-term commitments.

Reliable deployments and sustained customer demand would weaken the short thesis. Feldman also disclosed that he is long Cerebras stock, which matters when weighing his company’s rebuttal.

With the lock-up expired and the stock about 49% below its IPO price, November 12 puts potential share supply beside operating evidence. The next variable is whether revenue, margins and deployed capacity can settle the wafer-life dispute.