NEW YORK — Cerebras Systems, one of Wall Street’s most closely watched challengers to Nvidia in artificial intelligence computing, has fallen below its IPO price just months after one of the biggest semiconductor debuts ever, as investors confront a dangerous combination of Nvidia competition, heavy dependence on OpenAI and a wave of newly tradable insider shares.
Cerebras shares closed Friday at:
$166.43.
That left the stock nearly:
20% lower for the week
and below the company’s:
$185 IPO price.
The decline is even more dramatic compared with the excitement surrounding its May debut.
Cerebras shares opened their first day of trading at around:
$350
after pricing at $185.
The stock finished its first trading day above $300.
Now, less than five months later, more than half of that initial market enthusiasm has disappeared.
The question confronting investors is becoming much bigger than whether the stock simply rose too far after its IPO.
It is whether Cerebras can prove there is enough room beside Nvidia for a radically different AI computing architecture.
THE STOCK HAS LOST MORE THAN HALF ITS POST-IPO POP
Cerebras completed its IPO in May.
The company sold:
34.5 million shares
at:
$185 each.
That generated approximately:
$6.38 billion in gross proceeds.
The offering was one of the largest semiconductor IPOs ever.
Demand was intense.
When trading began on Nasdaq under the ticker:
CBRS,
shares opened around $350.
That immediate surge valued the company far above its private-market level and reflected enormous investor enthusiasm for anything connected to AI infrastructure.
But that enthusiasm has reversed quickly.
At Friday’s close of $166.43, Cerebras was:
below its IPO issue price
and dramatically below its first-day trading levels.
NVIDIA PRESSURE TRIGGERED THE LATEST SELL-OFF
The latest decline accelerated after research firm SemiAnalysis reported that OpenAI’s “Ultrafast” mode for its latest GPT-6.1 Sol model would run on:
Nvidia GPUs
rather than
Cerebras hardware.
The report has not been established as a cancellation of Cerebras’ broader relationship with OpenAI.
That distinction is critical.
But investors immediately focused on what the development could mean.
OpenAI is one of Cerebras’ most important customers.
If OpenAI decides Nvidia hardware is better suited for high-profile workloads that Cerebras expected to handle, Wall Street may have to reduce assumptions about Cerebras’ future revenue.
OPENAI IS CENTRAL TO THE CEREBRAS STORY
Cerebras announced earlier this year that it had entered a massive multi-year relationship with OpenAI.
The company described the agreement as involving:
750 megawatts of AI infrastructure
and said it was valued at:
more than $20 billion.
That made OpenAI one of the strongest pillars supporting the Cerebras investment thesis.
The logic was simple.
OpenAI needs enormous computing capacity.
Cerebras offers a radically different architecture designed for extremely fast AI inference.
If OpenAI deployed Cerebras broadly, the startup could scale revenue extraordinarily quickly.
That is why even an unconfirmed report suggesting Nvidia may take an important workload can move the stock sharply.
THIS DOES NOT MEAN OPENAI HAS WALKED AWAY FROM CEREBRAS
Investors need to separate several things.
A single workload running on Nvidia does not automatically mean:
the OpenAI-Cerebras contract has been canceled,
OpenAI has stopped using Cerebras,
or
the entire $20-billion-plus agreement is disappearing.
Cerebras previously said it was an OpenAI launch partner for GPT-5.6 Sol.
The company continues describing OpenAI as a major customer and partner.
The current concern is narrower:
Will Cerebras win as much future OpenAI inference traffic as investors previously expected?
That is still unresolved.
NVIDIA REMAINS THE INDUSTRY’S DOMINANT FORCE
Cerebras faces a competitor with extraordinary advantages.
Nvidia has built the most powerful ecosystem in AI computing.
Its strength is not only its GPUs.
It also controls a vast software ecosystem centered around:
CUDA
plus libraries and tools used by developers throughout the AI industry.
That creates a powerful lock-in effect.
Companies already building models on Nvidia hardware can often scale faster without redesigning their software stack for another architecture.
Nvidia also benefits from:
Massive production scale
Strong networking products
Developer familiarity
Cloud partnerships
and
Billions of dollars available for R&D.
Cerebras must convince customers that its speed advantage is powerful enough to overcome those advantages.
CEREBRAS IS NOT TRYING TO BUILD A NORMAL GPU
This is what makes the company interesting.
Cerebras does not simply make another graphics processor.
Its flagship hardware uses a:
wafer-scale processor.
Instead of cutting a silicon wafer into many small chips, Cerebras essentially turns most of the wafer into one enormous computing engine.
Its earlier WSE-3 processor contained approximately:
4 trillion transistors.
The design allows enormous amounts of data to move inside the system without repeatedly traveling between separate chips.
Cerebras argues this produces dramatically faster AI inference.
THE COMPANY SAYS SPEED IS ITS MAIN ADVANTAGE
Cerebras’ pitch is built around one idea:
fast AI is more valuable AI.
Inference is the process where a trained AI model actually answers users.
When a person asks a chatbot a question, generates code, analyzes financial information or uses an AI voice assistant, inference is happening.
Slow inference creates:
Long pauses
Lower productivity
and
Poorer user experience.
Cerebras believes extremely fast inference can unlock new applications where AI responses need to happen almost instantly.
These include:
Voice AI
Cybersecurity
Trading
Coding
and
Agentic AI systems.
THE NEW CS-4 IS SUPPOSED TO PUSH THAT SPEED ADVANTAGE FURTHER
Cerebras introduced its latest system:
CS-4
in August.
The company says the platform can deliver performance up to:
30 times faster
than GPU-based alternatives for some workloads.
Those are Cerebras’ own performance claims and can vary depending on model and workload.
But they illustrate the strategy.
Cerebras does not necessarily need to beat Nvidia at everything.
It needs to prove its architecture is dramatically better at specific high-value inference workloads.
If it can do that, customers may be willing to operate multiple types of hardware instead of relying exclusively on GPUs.
THE COMPANY’S BUSINESS IS STILL GROWING FAST
The stock decline comes despite strong reported operating growth.
In the second quarter, Cerebras reported:
$180.1 million in GAAP revenue
up:
74% year over year.
Its adjusted or “core” revenue reached:
$209.9 million
up:
103%.
Cloud and other services revenue reached:
$126 million on a GAAP basis.
That represented:
281% annual growth.
Core cloud revenue rose:
287%.
Those are extremely high growth rates.
CEREBRAS SAYS ITS CLOUD BUSINESS NEARLY QUADRUPLED
This matters because Cerebras increasingly wants to sell computing as a service rather than relying only on hardware systems.
Customers can access Cerebras computing through the cloud without buying an entire CS system themselves.
That creates recurring usage revenue.
It also makes Cerebras easier for:
Startups
Enterprises
and
AI developers
to adopt.
The company says demand for its cloud inference product has grown rapidly as customers seek lower latency.
$25.4 BILLION IN REMAINING PERFORMANCE OBLIGATIONS IS A HUGE NUMBER
Cerebras reported:
$25.4 billion
in remaining performance obligations.
That metric represents contracted revenue that has not yet been recognized.
In theory, it provides visibility into future revenue.
But investors need to understand an important distinction.
Remaining performance obligations do not equal:
cash already received
or
guaranteed near-term profit.
The revenue may be recognized over many years.
Contracts can include conditions.
And Cerebras still has to build and deploy the infrastructure required to deliver the service.
The backlog is impressive.
Execution will determine how much value it ultimately creates.
CEREBRAS EXPECTS REVENUE TO MORE THAN TRIPLE IN 2027
Management has laid out extremely aggressive growth expectations.
Cerebras says it plans to:
more than triple revenue in 2027.
That level of growth would be extraordinary for a semiconductor infrastructure company.
But it also raises expectations.
When a stock is valued around hypergrowth assumptions, anything that threatens a major customer can cause a severe reaction.
That is exactly what happened this week.
OPENAI CUSTOMER CONCENTRATION IS THE BIG RISK
Customer concentration is increasingly one of Wall Street’s biggest concerns.
A young infrastructure company may grow quickly because one major customer places an enormous order.
But that creates dependency.
If the customer:
reduces spending,
changes technology,
switches suppliers,
or
delays deployment,
the supplier can suffer dramatically.
OpenAI has enormous bargaining power because virtually every AI hardware supplier wants its business.
That includes:
Nvidia
AMD
Broadcom
Cerebras
and cloud giants developing their own chips.
OPENAI ITSELF USES MULTIPLE TYPES OF HARDWARE
There is another reason the competitive picture is complicated.
OpenAI is not necessarily choosing one chip provider for everything.
Large AI companies increasingly use:
multiple compute architectures
for different workloads.
One platform may be better for training.
Another may be better for high-speed inference.
Another may be cheaper for batch processing.
Another may have greater availability.
That means Nvidia winning one workload does not automatically eliminate Cerebras.
But it also means Cerebras cannot assume one large OpenAI agreement guarantees every future OpenAI workload.
NVIDIA HAS BEEN PUSHING AGGRESSIVELY INTO INFERENCE
Historically, Nvidia became famous for AI training.
But inference is now becoming an equally important market.
As hundreds of millions of people use generative AI tools every day, the amount of computing needed to generate responses is exploding.
Nvidia is therefore optimizing newer systems for:
Fast inference
Lower cost per token
and
Energy efficiency.
That brings it directly into Cerebras’ strongest claimed territory.
CEREBRAS IS ALSO TRYING TO DIVERSIFY BEYOND OPENAI
This may be the most important thing management can do.
Cerebras recently announced a major partnership with:
Gimlet Labs.
The cloud startup plans to deploy Cerebras CS-4 systems consuming roughly:
100 megawatts of power.
Deployment is expected over one to two years.
Gimlet plans to make Cerebras hardware available in its cloud environment beginning in:
2027.
Financial terms were not disclosed.
But the deal shows Cerebras is building relationships beyond OpenAI.
100 MEGAWATTS IS NOT A SMALL DEPLOYMENT
AI data centers consume enormous amounts of electricity.
A 100-megawatt deployment is substantial.
For comparison, a traditional data center might historically have consumed only a fraction of that amount.
The AI boom has radically changed infrastructure requirements.
Modern GPU and AI accelerator facilities increasingly require:
Hundreds of megawatts
and sometimes
Gigawatt-scale power planning.
That makes Cerebras’ Gimlet deal strategically meaningful even without a disclosed contract value.
CEREBRAS ALSO HAS AN AWS RELATIONSHIP
The company has announced a partnership with:
Amazon Web Services.
The goal is to make Cerebras’ high-speed inference technology easier to access through AWS environments.
That matters because AWS is the world’s largest cloud infrastructure provider.
Enterprise customers prefer infrastructure that integrates into systems they already use.
Partnerships with hyperscalers can therefore reduce barriers to adoption.
AMD IS ANOTHER CEREBRAS PARTNER — AND A COMPETITOR TO NVIDIA
Cerebras has also highlighted work with:
AMD.
This is another sign that the AI infrastructure market is becoming increasingly modular.
Companies do not necessarily need every component from one supplier.
A system can potentially combine:
Cerebras inference
AMD processors
Cloud infrastructure
and
Third-party networking.
That approach is sometimes described as:
disaggregated inference.
It is meant to reduce dependence on one vertically integrated vendor.
In practice, that vendor is often Nvidia.
BUT NVIDIA’S ECOSYSTEM IS EXTREMELY HARD TO DISPLACE
The challenge is that customers care about more than raw speed.
They also care about:
Software compatibility
Developer availability
Reliability
Supply
Integration
Support
and
Resale value.
Nvidia has spent years building strength in every one of these categories.
Even if a competing chip is faster in a benchmark, a customer may still choose Nvidia because migrating software is expensive or risky.
This is one reason Nvidia’s dominance has been so durable.
CEREBRAS’ SECOND PROBLEM THIS WEEK WAS THE LOCKUP EXPIRATION
Competition was not the only thing pushing the stock lower.
Another technical factor hit at the same time:
post-IPO lockup restrictions began easing.
IPO lockups prevent many insiders, employees and early investors from immediately selling their shares after a company goes public.
The goal is to prevent a flood of new supply immediately after listing.
When those restrictions end or partially expire, additional shares become eligible to trade.
That can pressure the stock even when the company’s underlying business has not changed.
WHY LOCKUPS MATTER SO MUCH
Imagine there are relatively few Cerebras shares available for public trading.
Demand rises.
The stock can move sharply higher.
Then millions of previously restricted shares suddenly become eligible for sale.
The supply of potentially tradable stock increases.
Even if only a fraction of shareholders actually sell, investors may anticipate more selling pressure.
That expectation itself can push prices down.
This is a common pattern after high-profile IPOs.
CEREBRAS WARNED ABOUT THIS BEFORE THE IPO
Cerebras explicitly disclosed the risk in SEC filings.
The company said insiders and other shareholders were subject to lockup and market-standoff restrictions.
Its filings estimated that as many as approximately:
171.1 million shares
could become released from various restrictions during the broader lockup period under early-release provisions.
That does not mean 171 million shares suddenly became tradable this week.
The releases occur under different conditions and schedules.
But it demonstrates how large the pool of previously restricted stock is compared with the shares sold in the IPO.
INSIDER SALES ADDED TO THE NERVOUSNESS
SEC filings also show Cerebras executives filing transactions around the same period.
Some were made under:
Rule 10b5-1 trading plans.
These plans are generally established in advance and allow corporate insiders to schedule stock sales according to preset conditions.
That means an insider sale is not automatically evidence that the executive suddenly became bearish on the company.
But when insider shares hit the market while the stock is already falling, investor psychology can become much worse.
THE IPO ITSELF SOLD ONLY 34.5 MILLION SHARES
This helps explain why unlocks matter.
Cerebras sold:
34.5 million shares
in the IPO.
Yet its filings show a much larger number of shares held by:
Founders
Employees
Venture investors
and
Early shareholders.
As those shares gradually become tradable, the public float can expand dramatically.
That is healthy for liquidity over the long term.
In the short term, it can weigh on the stock.
WALL STREET IS NOW TESTING THE IPO VALUATION
At the $185 IPO price, investors were effectively paying for enormous future growth.
The first-day rally above $300 pushed expectations even higher.
That valuation assumed Cerebras would become one of the most important challengers to Nvidia.
For that to happen, the company likely needs to:
Deliver rapid revenue growth
Convert backlog into revenue
Diversify customers
Improve profitability
and
Keep its technological advantage.
The lower stock price suggests Wall Street is now demanding stronger evidence before paying the same premium.
THIS DOES NOT MEAN THE AI CHIP BOOM IS OVER
The Cerebras selloff is happening while AI infrastructure spending continues exploding.
Amazon, Microsoft, Meta, Google, OpenAI, Anthropic and others are spending hundreds of billions of dollars on:
Data centers
AI chips
Networking
and
Power infrastructure.
Nvidia remains one of the biggest beneficiaries.
Broadcom is gaining through custom chips.
AMD is expanding.
And multiple startups are attempting to carve out specialized markets.
The total market continues to grow.
The problem for Cerebras is not a lack of demand.
It is proving that enough of that demand will flow specifically to Cerebras.
THE AI CHIP MARKET MAY NOT BE WINNER-TAKE-ALL
There is an argument in Cerebras’ favor.
AI workloads are growing so rapidly that the market may support multiple architectures.
A massive AI company could use:
Nvidia GPUs for training,
Cerebras for ultra-fast inference,
custom ASICs for specialized workloads,
and
AMD accelerators for additional capacity.
This is already beginning to happen.
Large customers increasingly want to avoid dependence on any single supplier.
That creates an opening for Cerebras.
NVIDIA’S SIZE CAN ACTUALLY HELP ALTERNATIVES
Nvidia’s dominance has created motivation for customers to diversify.
Customers worry about:
Supply shortages
Pricing power
and
Vendor lock-in.
That gives alternatives a reason to exist even if Nvidia remains technologically strong.
Cerebras does not necessarily need to dethrone Nvidia.
It needs enough customers to believe that combining Cerebras with other architectures improves:
Speed
Cost
or
Resilience.
That is a more realistic competitive target.
PROFITABILITY STILL MATTERS
Fast revenue growth can distract from the financial fundamentals.
Cerebras remains in a capital-intensive business.
The company must fund:
Chip manufacturing
Systems
Data-center infrastructure
Cloud capacity
Engineering
and
Software development.
Scaling 10 times faster requires enormous spending.
That is one reason Cerebras raised more than:
$6 billion
in its IPO.
The money gives it substantial resources.
But investors ultimately need to see those investments produce durable profits.
DATA CENTER CAPACITY IS EXPANDING MORE THAN TENFOLD
Cerebras says its manufacturing capacity is expected to scale more than:
10 times during 2026.
It also reported approximately:
600 megawatts of data-center capacity under contract
as of Q2.
Those numbers show the scale management is preparing for.
But infrastructure commitments work both ways.
If demand arrives, that capacity can produce enormous revenue.
If demand falls short, expensive infrastructure can become a financial burden.
Execution is critical.
POWER IS BECOMING AS IMPORTANT AS CHIPS
The AI industry is increasingly constrained not by demand for computing but by:
Electricity
Transformers
Data centers
Cooling
and
Grid connections.
A company can have a brilliant chip architecture and still struggle if it cannot deploy enough powered infrastructure.
Cerebras’ hundreds of megawatts of contracted capacity are therefore strategically important.
But they also expose the company to the broader race for energy infrastructure.
CEREBRAS HAS ONE MAJOR ADVANTAGE: DIFFERENT ARCHITECTURE
Most Nvidia competitors attempt to build better conventional accelerators.
Cerebras took a much more radical approach.
That creates higher technological risk.
But it also creates the possibility of genuine differentiation.
If wafer-scale computing proves materially faster and more efficient for large AI inference workloads, Cerebras could occupy a category that Nvidia cannot easily eliminate simply by lowering GPU prices.
The company’s future depends heavily on proving that advantage under real commercial workloads.
THE BIGGER STORY: CEREBRAS DOES NOT HAVE TO BEAT NVIDIA — BUT IT MUST PROVE OPENAI IS NOT ITS ENTIRE STORY
Cerebras entered the public market as one of the boldest bets on a post-Nvidia AI world.
It raised:
$6.38 billion.
Its shares initially exploded above:
$300.
It announced a relationship with OpenAI worth more than:
$20 billion.
Its cloud business nearly quadrupled.
And it says contracted future business now exceeds:
$25 billion.
Yet five months after the IPO, the stock has fallen below:
$185.
Why?
Because Wall Street is suddenly focusing less on the size of the AI opportunity and more on how fragile individual winners can be.
One report suggesting Nvidia won a high-profile OpenAI workload helped erase billions of dollars in market value.
At the same time, post-IPO lockups are releasing additional stock into the market.
Neither issue proves Cerebras’ technology has failed.
But together they expose the company’s central vulnerability:
too much of the investment story still depends on investors believing Cerebras will capture a meaningful share of workloads from Nvidia—and that OpenAI will remain one of its biggest supporters.
The company now has to prove that its 600 megawatts of contracted capacity, its rapidly growing cloud business and partnerships beyond OpenAI can build a broader customer base.
Cerebras does not need to defeat a $5-trillion Nvidia empire to succeed.
But after falling below its IPO price, it now has to prove that being dramatically faster at AI inference is enough to build a durable business beside it.
