When a Wall Street IPO prices at $185 and opens at $350, that’s normally a signal. Sometimes a good one, often a suspicious one. Cerebras Systems (CBRS) delivered both yesterday: a 68 percent spike on its first trading day, a high of $350, a close at $311, and a slight pullback in this morning’s pre-market.
For AI investors, Cerebras is the hottest newcomer since the Nvidia rally. The company positions itself explicitly as an alternative to Nvidia’s H200/H100 – with different architecture, different pricing model, different market. The question everyone is asking now: is this the next real AI chip story, or is Cerebras just riding the FOMO wave of Trump-Xi euphoria?
Both answers are possible. Let’s look at the data honestly.
What Cerebras Actually Builds
Cerebras Systems was founded in 2016 in Sunnyvale. The company builds Wafer-Scale Engines (WSE) – AI chips that don’t come in classic semiconductor format, but as gigantic single-wafer chips. While an Nvidia H100 is about 814 mm² large, a Cerebras WSE-3 has an area of 46,225 mm². That’s 57x larger.
What does that give you? In AI training and inference, it’s about data transfer between compute cores. The closer the cores sit, the faster the model runs. Nvidia must connect thousands of H100 chips via NVLink. Cerebras has all 900,000 cores on a single wafer – no external networking needed.
In theory, that’s spectacular. In practice, Cerebras has only sold about 100 productive installations worldwide since 2016. Nvidia delivers more H100 systems weekly than Cerebras has in 9 years total. The size difference isn’t 10x, isn’t 100x – it’s 1000x.
Why 68 Percent on Day One Anyway
Three reasons. None of them is “Cerebras will topple Nvidia,” even if some retail investors believe that right now.
First, IPO market mechanics. Cerebras was priced at $185 at a valuation of roughly $7-8 billion. At that pricing, the float (available shares) was small, demand books oversubscribed. That’s classic IPO underpricing: lead investment banks price low so the stock pops on day one and allocation recipients make profits. The +68 percent says nothing about the company.
Second, AI investment hype. Anthropic closed a $30 billion funding round at a $900 billion valuation today. In this climate, every capital that even remotely smells like “AI chip story” flows into the next available stock. Cerebras was available at the right time.
Third, fund diversification needs. Many funds can’t hold more than 5-7 percent in a single stock for risk management. The “AI trade” allocations are extremely concentrated toward Nvidia. Cerebras gives these funds a way to diversify AI exposure – even if the fundamental story is thinner. To understand the bigger picture, see our AI Stocks Investing 2026 hub, where we map the entire value chain from TSMC to Cerebras.
What the Cerebras Story Really Is
Honestly: Cerebras has real technology. Wafer-scale isn’t marketing, it’s engineering. The WSE-3 can serve certain workloads faster than an H100 cluster – especially Large Language Model inferencing with long context windows. That’s real.
But: Cerebras has a massive scaling problem.
First, a Cerebras system costs millions per unit. Nvidia can produce profitably for hyperscalers like AWS, Azure, and GCP because volumes are massive. Cerebras essentially sells custom hardware to research labs and individual enterprise customers. That’s a different business model – more like a Cray supercomputer than an Nvidia GPU seller.
Second, Cerebras depends on a single customer to an unhealthy degree. G42 from the United Arab Emirates is Cerebras’ largest customer with an estimated 80 percent or more of revenue in the past two years. If G42 chooses another supplier, Cerebras revenue collapses.
Third, the software ecosystem is extremely thin. Nvidia’s success is based not just on better GPUs – it’s based on CUDA, the standard framework for AI development worldwide. Anyone training an AI model today writes CUDA code. Cerebras has its own framework that almost nobody outside the company uses. That’s a massive adoption hurdle that money alone can’t solve.
Who Wins With Cerebras – And Who Loses
In the first week after IPO, pre-IPO investors and insiders who entered at $185 or below win. They have enormous paper gains now. At the current price of around $311 (Thursday close), Cerebras’ valuation has risen to roughly $12-14 billion.
In the second and third week, retail investors who entered today or yesterday at $300+ will likely lose. The typical post-IPO pattern: spike on day one, slight pullback in week 2-4, then massive drop when the lock-up period (typically 180 days) ends and insiders can sell.
Who really benefits from the Cerebras story are:
- TSMC, which produces the wafers for Cerebras
- Hyperscaler competition, which gets more negotiating power versus Nvidia through Cerebras
- Anthropic, OpenAI and other AI labs, which will see lower inference costs through chip market competition
The question “can Cerebras attack Nvidia?” is the wrong question. The right question is: “Will the AI compute market be homogeneous or heterogeneous in 5 years?” If the answer is heterogeneous (which is likely), there’s room for Cerebras alongside Nvidia, AMD, Intel, and Chinese providers. But that room is smaller than Cerebras’ current valuation suggests.
What BMI Readers Can Do With This
For smart money followers, Cerebras is a perfect case study. We can track in the next 90 days which institutional investors enter and exit. The next 13F filing will show whether Tiger Global, Coatue, Ark Invest or other large tech funds enter.
For active traders, Cerebras is volatile playing material. +68 percent on day one, probably -15 percent in 2 weeks, then sideways with high volatility until the first earnings season. That’s trader land, not investor land.
For buy-and-hold investors with 5+ year horizons, Cerebras is too expensive today. At $12-14 billion valuation with single-customer risk, thin software ecosystem, and massive competitive disadvantage versus Nvidia, the risk-reward is unattractive. Wait at least 6 months until after earnings, then reassess.
Three Lessons from the Cerebras IPO
First: AI hype isn’t AI business. A 68 percent spike on day one means IPO banks priced conservatively. It doesn’t mean the stock is underpriced.
Second: Competing with Nvidia is harder than it looks. Nvidia has not only better hardware – Nvidia has the CUDA ecosystem, decades of developer relationships, and scale advantages that Cerebras won’t catch up in 5 years.
Third: Diversification in AI is real – but not through Cerebras. If you want AI exposure beyond Nvidia, TSMC, ASML, Broadcom, and Marvell are the more structural plays. Cerebras is a story, not a foundation.
Sam Stovall of CFRA said this week that bull markets die from mispriced risk, not from geopolitics. Cerebras is exactly such mispriced risk: spectacular story, thin business, high valuation. It’s the kind of stock that generates hype in bull markets and loses 50-70 percent in bear markets.
Whoever understands that can decide whether to play along as a trader – or whether the money is better spent on Microsoft, TSMC, or Nvidia.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
