Cerebras shares plummet, mixed quarterly results test AI narrative

Shares of artificial intelligence company Cerebras Systems fell 17% in after-hours trading on Wednesday, although the company exceeded Wall Street expectations on most measures and raised its full-year revenue forecast. This market reaction is of great significance to investors in the artificial intelligence infrastructure industry and helps assess whether leading AI infrastructure companies can continue to maintain high valuations.

Revenue exceeded expectations, losses narrowed

According to financial report data, Cerebras (Nasdaq: CBRS) core revenue in the second quarter reached US$210 million, doubling from the same period last year, exceeding analysts 'expectations of US$191 million. After adjusted, the company lost 5 cents per share, well below the expected loss of 17 cents per share; the adjusted operating loss was $34 million, also significantly below the forecast of $63 million. Based on U.S. General Accounting Standards (GAAP), total revenue was US$180.1 million, a year-on-year increase of 74%.

Wafer level engine encounters setbacks

Cerebras is known for its Wafer-Scale Engine, a design that retains an entire 300-mm silicon wafer as a single giant processor, built for artificial intelligence and scientific computing tasks. The hardware earned it the label of a "Nvidia competitor," but in the second quarter, the business took a hit. Physical hardware revenue fell 23% year-on-year to $54.1 million. CEO Andrew Feldman said this was due to delivery scheduling issues rather than weak demand. "There will be fluctuations in the hardware business," he said, explaining that it is sometimes impossible to confirm sales because customers lack data center space to install machines the size of Cerebras equipment.

Cloud services become new engine

Based on GAAP, cloud and other services revenue reached US$126 million, a year-on-year increase of 281%; core cloud revenue climbed 287% to US$127.7 million. This means that the company, which prides itself on challenging Nvidia with chips, has now become leasing computing power rather than selling hardware. Cerebras also turned into a GAAP net loss of $450.5 million (approximately $2.89 per share), compared to a profit of $309.5 million in the same period last year. The company attributed the reversal to $386.6 million in equity incentive spending related to its May initial public offering. Core gross profit margin increased by nearly 940 basis points from the second quarter of 2025 to 41%, thanks to premium pricing for fast reasoning services.

Performance guidance increased

Management raised its expectations with the help of earnings: Cerebras increased its core revenue guidance for fiscal 2026 from $855 million to $865 million to $880 million to $890 million, higher than analysts 'expectations of $867.6 million. Core revenue guidance for the third quarter is US$214 million to US$216 million, higher than the market consensus of approximately US$212 million, and gross margins are expected to be 38% to 40%. Chief Financial Officer Bob Comin said the company plans to more than triple revenue in 2027 as production scales expand and component costs fall. CEO Feldman said in a statement: "Speed changes the scope of AI's capabilities," adding that "this opens up entirely new markets." He also described gross margins as "in good shape and growing because rapid reasoning service pricing enjoys a premium."

Post-IPO balance sheet situation

Cerebras 'initial public offering in May left sufficient funds and the company totaled US$6.4 billion in total earnings. As of the end of June, the company also had $8.6 billion in cash, cash equivalents, restricted cash and short-term investments, and $25.4 billion in remaining performance obligations. During regular trading on Wednesday, the stock closed at $262.06, about 42% above its $185 IPO price, before an after-hours decline erased some of the gains. The company is also expanding partnerships to expand its influence, including a partnership with AMD and an arrangement that allows OpenAI to run models such as GPT 5.6-Sol on its systems.