Cerebras Systems (CBRS) is up nearly 20% intraday on June 8, trading around $240 — off a previous close of $201.01 and inside a daily range of roughly $204 to $247. That is a wide range. It is also, in this market, an entirely rational one for a stock still in active price discovery less than a month after its IPO.

Here is where I am at with this. The move is real, the volume is heavy, and the underlying business metrics are not fiction. But traders chasing this without understanding what they are actually buying could find themselves holding a very expensive bag in a very illiquid stretch. Both things are true at once.

The IPO Context

Cerebras went public on May 14, 2026, pricing 30 million Class A shares at $185 — the largest U.S. tech listing since Uber in 2019. The stock opened at $350, closed its first session at $311.07, a roughly 68% gain from the IPO price. Within days it hit an all-time high of $386.34. It then sold off hard into the low-$200s over the following weeks, touching $196.73 on June 5 before today’s snapback.

That is a textbook post-IPO cycle: euphoric open, profit-taking from early allocations, and then a tug-of-war between dip buyers and longs who got in near the highs. The current move looks like the first real technical bounce off what may have been the post-IPO floor.

The Business — What the Numbers Show

Cerebras reported 2025 revenue of $510 million, up 76% year-over-year from $290.3 million in 2024. The company grew from $24.6 million in revenue in 2022 to $510 million in 2025 — more than a tenfold increase in three years. That growth rate is genuinely exceptional. GAAP net income came in at $237.8 million for 2025, a dramatic swing from a $481.6 million loss in 2024.

The part people skip: that GAAP profit includes a one-time non-cash gain of roughly $363.5 million. Strip that out and the GAAP operating loss was $145.9 million in 2025. The company also carries a revenue backlog of $24.6 billion as of December 31, 2025 — a number that is impressive in scale but contingent on foundry access the company does not formally control. Cerebras buys wafers from TSMC on a purchase-order basis, with no long-term supply commitment. That is a meaningful supply-chain risk embedded in an otherwise bullish backlog figure.

Customer concentration is the other issue worth flagging. Approximately 86% of 2025 revenue came from two UAE-linked entities — G42 and MBZUAI. U.S.-billed revenue actually declined 34% in 2025. The company is actively working to diversify: OpenAI signed a multi-year agreement valued above $20 billion in January 2026, and AWS formalized a term sheet in March 2026 to deploy Cerebras hardware in its data centers. Those are real catalysts. But they are forward-looking, and the current revenue base remains heavily concentrated.

The Technology Edge

This is where Cerebras earns serious attention from engineers, not just traders. The WSE-3 — Wafer Scale Engine 3 — is the largest AI chip ever commercially produced. It packs 4 trillion transistors and 900,000 cores onto a single silicon wafer using TSMC’s 5nm process, with 44GB of on-chip SRAM delivering memory bandwidth of 21 petabytes per second. The WSE-3 die is 57 times larger than Nvidia’s H100.

What that architecture actually solves is a latency and memory-bandwidth problem. Nvidia GPU clusters rely on NVLink and PCIe for chip-to-chip communication — those remain off-die and inherently higher latency. The WSE-3 implements on-wafer interconnects across its mesh of cores, eliminating that bottleneck. On low-batch inference workloads — specifically Llama 3 70B at batch size 1 — Cerebras claims 2,100 tokens per second versus roughly 500 to 700 tokens per second on a single H100.

The counterargument: at larger batch sizes, H100 closes much of that gap on throughput-per-dollar. And Nvidia’s CUDA ecosystem remains the deepest software moat in enterprise AI. Cerebras uses a proprietary SDK. If you migrate to it, migration back requires a full rewrite of your inference client. That is a real friction point for enterprise buyers who have spent years embedding CUDA workflows.

Valuation and Analyst Positioning

At roughly $240, CBRS trades at a market cap around $53 billion against $510 million in 2025 revenue. That implies a price-to-sales multiple north of 100x. The trailing P/E using reported GAAP earnings is over 590x — a number that loses meaning quickly when the underlying profit includes a non-recurring $363 million gain. On a GAAP operating basis, the company lost money in 2025.

Nine analysts currently cover CBRS with an average Buy rating and a 12-month consensus price target of $290.56. Wedbush initiated with Outperform and a $270 target, Rosenblatt and Needham both initiated at Buy, and Barclays tagged it Overweight with a $280 target. Those targets are roughly 20% above current levels — not dramatic upside given the volatility profile of the stock.

For context, Nvidia trades at approximately $204 with a market cap approaching $5 trillion and trailing revenue of $253.5 billion. Nvidia still holds roughly 90% of the AI accelerator market. Cerebras is not replacing that — not this year, probably not in this decade on a broad basis. What it can do is carve a defensible niche in inference workloads where memory bandwidth is the constraint.

Levels and Risk Framework

The 52-week range runs from $185 to $386.34. The IPO price of $185 and the June 5 low of $196.73 are the logical reference points for risk management. Today’s move has the stock pressing toward the $240 to $248 zone — a range that has acted as interim resistance in the recent trading history. A clean hold above $240 with sustained volume would be constructive. Failure there likely puts the $210 to $215 area back in focus.

What matters most for positioning: CBRS has a daily average volume of roughly 5 million shares. Today’s session is running above that average, which confirms institutional participation in the move — not just retail chasing. The balance sheet is solid, with more than $1.1 billion in cash and short-term investments against total liabilities under $1 billion. The company is not a liquidity risk in the near term.

The scenario that works against longs: customer concentration does not diversify fast enough, TSMC supply tightens during a period of peak demand, and the OpenAI and AWS revenue ramp takes longer than the backlog number implies. In that environment, a stock trading at 100x revenue with no GAAP operating profit has very little technical support outside of the IPO price level.

Slight tangent, but it matters: the broader AI hardware theme is getting increasingly crowded at the public-market level. Marvell, Broadcom, and AMD are all competing for hyperscaler custom silicon spend. Cerebras is differentiated by architecture — but differentiation alone does not protect a valuation multiple.

What this move is doing today is giving traders a clean, high-beta expression of AI hardware enthusiasm in a name that has real revenue, a genuine architectural edge, and a risk profile that rewards preparation over conviction. The data is there. The questions around customer concentration, foundry access, and forward profitability are also there. Neither cancels the other out. Both belong in your framework before size gets added.

Watch the $240 level. Watch volume into the close. The next earnings report is scheduled for September 2, 2026.

For informational and educational purposes only. Not investment advice. Trading involves risk, including loss of principal.