After the debut surge of "Nvidia Challenger" Cerebras, Cathie Wood quickly built a position

According to the latest trading disclosure information obtained by Zhitong Finance APP, "Woodstock" Kathy Wood's ARK Innovation ETF (ARKK) and ARK Next Generation Internet ETF (ARKW) purchased 105,616 shares of newly listed AI chip manufacturer Cerebras Systems (CBRS.US) on Thursday. Based on Wednesday's closing price, the combined position is valued at approximately $32.8 million.

Silicon Valley AI chip company Cerebras Systems officially landed on NASDAQ on May 14th, Eastern Time, staging a long-awaited tech IPO frenzy.

The company's IPO was priced at $185 per share, significantly up from the initial expectation of $115 to $125. In the week leading up to the listing, due to overwhelming institutional demand—subscription orders exceeded the number of shares available for issuance by 20 times—Cerebras raised the offering price twice and expanded the sale scale. After opening, the stock price surged to $350, reaching a peak of $386 during the session, an increase of 108%, triggering NASDAQ's volatility circuit breaker. It ultimately closed at $311.07, up 68.2% from the offering price.

At the closing price, Cerebras has a market capitalization of approximately $67 billion; if options and warrants are included, the fully diluted valuation is about $86 billion. The company sold 30 million shares in this offering, raising $5.55 billion, and if underwriters exercise the 4.5 million share over-allotment option, the total financing scale could reach $6.38 billion. This is the largest IPO globally since 2026 and the largest IPO in the U.S. tech industry since Uber's listing in 2019. The number of subscriptions attracted by this IPO exceeded 20 times the number of shares issued.

Wafer-Level Chips: Disruptive Technology or Expensive Artwork?

This is not an isolated position-building action. On the same day, ARK reduced its holdings in Taiwan Semiconductor Manufacturing Company (TSM) by 41,540 shares (approximately $16.6 million), AMD by 11,510 shares (approximately $5.13 million), and semiconductor testing equipment manufacturer Teradyne by 22,576 shares (approximately $8.2 million). Some analysts described this portfolio adjustment as a "generational hardware swap"—moving funds from mature semiconductor giants to bet on emerging computing infrastructure that challenges the existing landscape with wafer-level chip (WSE) technology.

For Kathy Wood, who consistently advocates for a "disruptive innovation" investment philosophy, this transaction is a typical projection of her 2026 investment logic—she clearly stated at the beginning of the year that a new round of large investment cycles combined with deregulation trends would create significant benefits, with core themes including AI, robotics, computing power, and new energy. The heavy investment in Cerebras on its first day of listing demonstrates her firm bet on the long-term prospects of the AI infrastructure sector Cerebras has attracted attention primarily due to its unique technological path. The company focuses on WSE (Wafer-Scale Engine) technology, which directly manufactures a single AI processor from an entire 12-inch wafer, rather than cutting multiple independent chips from the wafer like traditional manufacturers. Its third-generation product, WSE-3, uses a 5nm process, has a chip area of 46,225 square millimeters, integrates 900,000 computing cores and 40 trillion transistors, and achieves AI computing power of 125 petaflops.

Compared to traditional GPU clusters, the area of WSE-3 is approximately 58 times that of NVIDIA's B200, with on-chip memory bandwidth 2,625 times greater than the latter, and its inference speed is reportedly 10 to 20 times faster than mainstream GPU solutions.

CEO Feldman stated that a real turning point will occur in the first half of 2025: "By then, AI models will be smart enough to have real practical value. Our business exploded last year." In his view, as AI transitions from the training era to the inference era, the WSE architecture, which can generate tokens at a lower cost and faster speed, will encounter structural opportunities.

However, technological advantages do not guarantee commercial success. Analysts point out that once a wafer-scale chip has process defects, it cannot simply scrap a small portion of the chip; the entire wafer must be discarded, which imposes natural constraints on production flexibility and yield. Even if Cerebras claims to have developed a fault-tolerant architecture that can bypass defective blocks, the stability of large-scale production remains to be verified. Additionally, NVIDIA is expected to maintain an 86% market share in the AI accelerator market in 2025, making its position difficult to shake in the short term.

The Financial Truth of Cerebras: Accounting Magic Behind Profitability

However, behind this capital feast, Cerebras's financial data hides concerns overshadowed by market exuberance.

The prospectus reveals that Cerebras's revenue in 2025 is projected to be $510 million, a 76% year-on-year increase, and it has successfully "turned a profit"—with a net profit of $238 million and earnings per share of $1.38, a significant improvement from a loss of $9.90 per share the previous year. Looking at these numbers alone, a high-growth and profitable AI chip company emerges vividly.

But a closer examination of the accounting records reveals a story far more complex than it appears. In 2025, Cerebras recorded a GAAP net profit of $237.8 million, but $363.3 million of that came from a one-time non-cash accounting adjustment—resulting from the termination of a forward contract liability related to the UAE AI company G42, which generated a book gain. Excluding this item and adding back stock-based compensation costs, the company's non-GAAP net loss for 2025 would be $75.7 million, even widening from a loss of $21.8 million in 2024.

This means that Cerebras's "profitability" is essentially the result of an accounting treatment rather than the realization of operational profits. In an IPO with a valuation soaring to nearly $70 billion, this financial detail warrants caution from every investor The Dual Dilemma of G42's Shadow and OpenAI Dependency: Customer Concentration

Cerebras faces another core risk in the extreme concentration of its customer structure. In 2024, over 85% of Cerebras' revenue came from the UAE AI company G42. This high dependency directly triggered a national security review by the Committee on Foreign Investment in the United States (CFIUS), leading to Cerebras withdrawing its IPO application in 2024. The updated prospectus shows that G42's revenue share dropped to 24% last year, but the Mohamed bin Zayed University of Artificial Intelligence in the UAE still contributed 62% of the revenue. The two Middle Eastern entities combined still account for 86% of the company's revenue in 2025.

CEO Andrew Feldman responded by stating, "There are indeed some whale-level customers in the market, which is one of the characteristics of this market." However, rather than being a market characteristic, it is more of a business model issue— a company benchmarking against Nvidia should ideally sell chips to "everyone," rather than just a few sovereign-related customers.

To address this concern, Cerebras has been actively expanding its domestic customer base in the United States over the past year. In January 2026, the company signed a multi-year collaboration agreement worth over $20 billion with OpenAI, which will deploy 750 megawatts of high-speed AI inference computing power for OpenAI, along with a $1 billion loan and warrant arrangement. In March, AWS announced it would deploy Cerebras CS-3 chips in its data centers, making them available to developers through Amazon Bedrock.

However, the shift from reliance on G42 to OpenAI is essentially "replacing one dependency with another." According to Morgan Stanley's forecast, 90% of Cerebras' sales in the next two years are expected to come from OpenAI. The market can always defend concentration with the argument that "large customers are quality customers," until the day the large customer stops placing orders.

Valuation Debate: Is a 100x Price-to-Sales Ratio Justifiable?

With a market capitalization of $67 billion, Cerebras has a price-to-sales ratio of as high as 130 times based on projected revenue of $510 million in 2025. Is this valuation reasonable for a company that is still losing money on a non-GAAP basis? There is significant divergence in market opinions.

A senior research analyst at Renaissance Capital pointed out that if calculated based on an IPO price of $185 and projected revenue and EBITDA for 2028, the valuation is still reasonable; however, measured against the current trading price, the long-term valuation is already high. One analyst bluntly stated, "The market is filled with fervent fantasies about AI, but fervor is often not a rational pricing tool."

However, Cerebras' remaining performance obligations (RPO) amount to $24.6 billion, including over $20 billion in multi-year contracts with OpenAI, which supports its high growth expectations. If these contracts can be successfully delivered and revenue realized, the current valuation is not entirely without fundamental support. But the premise is—AI inference demand indeed materializes as Feldman predicts "Exploding 1 million times," and Cerebras' wafer-level chips can maintain a technological advantage in this computing power competition