Cerebras stock dropped after its public debut and stayed down for a long time, though it has started to bounce back lately. The company makes massive, powerful chips for artificial intelligence, but investors are currently worried about potential legal troubles and how the business will perform as the hype around AI cools down.
What does it do?
Cerebras Systems is a hypergrowth semiconductor and AI infrastructure company that earns money by selling specialized high-performance computer systems and cloud-based AI training services. Unlike traditional chipmakers that cut small chips from a silicon wafer, Cerebras uses the entire wafer to create a single giant processor called the Wafer-Scale Engine. Customers pay for these CS-3 systems as physical hardware for their data centers or subscribe to the company's cloud service to train their AI models on Cerebras hardware. This model provides both large upfront hardware sales and steady, recurring service fees.
Where does revenue come from?
The majority of revenue comes from high-end hardware sales, though AI cloud services are the fastest-growing part of the mix. The company splits its business into Hardware, which includes the sale and maintenance of the CS-3 systems, and Cloud and Other Services, where customers rent compute time on Cerebras supercomputers. Geographic data is limited, but a significant portion of current revenue is driven by large-scale supercomputer builds in the Middle East through partnerships like G42.
Who are its customers?
Cerebras Systems serves sovereign governments building national AI infrastructure, large-scale cloud providers, and enterprise companies training their own massive AI models. In the most recent quarter, the company reported total GAAP revenue of $193.4 million, a 93% increase over the $100 million generated in the same period last year. A critical portion of this revenue is concentrated in massive "Condor Galaxy" supercomputer projects, which are multi-phase builds involving thousands of systems. The company is actively working to diversify its base by attracting more enterprise clients to its cloud-based "inference" and training services.
What gives it staying power?
Cerebras has staying power because its wafer-scale integration is technically extremely difficult for competitors to replicate, creating a proprietary hardware advantage. By keeping all computing on a single piece of silicon, it avoids the speed limits of traditional networking. This creates high switching costs because once an AI model is optimized for Cerebras silicon and its "CSoft" software stack, moving to a different system requires significant re-engineering.
Where is it headed?
Cerebras is making a major strategic bet on becoming the primary alternative to Nvidia for training the world's largest AI models. Management is focused on scaling its "Condor Galaxy" supercomputer network to provide the massive compute power needed by national governments and hyperscale clouds. If successful, Cerebras will transition from a niche hardware provider to a global backbone for the next generation of AI infrastructure.
Revenue is accelerating sharply as the company scales production of its latest generation of hardware. Total revenue grew 93% year-over-year in the most recent quarter to $193.4 million, driven by massive demand for AI training clusters. This acceleration shows that the company is successfully moving from prototype stages to large-scale commercial deployments.
Cash generation is currently negative as the company pours capital into building its own cloud supercomputers. Free cash flow was negative $390 million last year, reflecting the heavy cost of manufacturing the world's largest chips before they are sold or deployed for services. This gap is expected for a company in this stage of hypergrowth, but it requires a solid cash cushion to sustain.
The balance sheet is managed conservatively with a low debt-to-equity ratio of 0.49x. This level of leverage is appropriate for a high-growth tech company, providing enough flexibility to fund research without over-burdening the business with interest payments. The company maintains sufficient liquidity to fund its current ramp-up toward profitability.
Cerebras is a hypergrowth business that is rapidly approaching a major financial turning point as it moves toward profitability.
Revenue growth of 93% year-over-year proves that there is massive, realized demand for alternatives to standard GPU clusters. The company also reached positive Adjusted EBITDA of $12.7 million this quarter, showing that the underlying business model becomes profitable as it reaches scale.
The heavy reliance on a single major partner, G42, for a large portion of the revenue backlog remains the biggest risk. If this partnership were to slow down or face regulatory hurdles, the company's path to profitability would be significantly delayed.
The AI semiconductor market is approximately $100 billion today and is growing at roughly 35% annually, putting it on track to exceed $350 billion by 2030. While pricing power is currently high due to a global shortage of high-end chips, it is structurally shaped by a "winner-takes-most" dynamic where the fastest and most efficient architecture attracts the most capital. Cerebras stands as a high-performance challenger that occupies the most demanding tier of the market, offering a specialized alternative to the standard cluster-based approach.
The competitive dynamic is dominated by a single giant, Nvidia, which has created a rational but difficult environment for challengers. Barriers to entry are astronomically high due to the hundreds of billions of dollars required in R&D and the deep software lock-in of existing systems. Long-term pricing power depends entirely on whether a challenger can provide a performance gain large enough to justify the cost of switching software stacks.
Nvidia represents the most dangerous threat because its CUDA software platform is the industry standard that almost all AI developers already know how to use. Broadcom threatens the business by enabling the largest cloud companies to build their own custom chips, potentially shrinking the market for independent chip makers. Start-ups like Groq compete by focusing on specific parts of the AI lifecycle, such as "inference," where they may offer better speed than a general-purpose system.
Cerebras is currently gaining share in the elite segment of the market, evidenced by its 93% revenue growth and massive supercomputer contracts.
The primary source of protection is the company's proprietary Brand & IP, specifically the patents and manufacturing secrets required to build a functioning wafer-scale processor. By integrating the entire wafer into one chip, Cerebras achieves a physical speed advantage that clusters of smaller chips cannot match. This hardware lead is proven by the company's ability to win multi-billion dollar contracts against much larger incumbents.
The current GAAP gross margin of 44.6% and the recent move to positive Adjusted EBITDA suggest that the business is starting to capture the value of its technical edge. While these numbers are lower than Nvidia's, they are consistent with a company that is still in the heavy investment phase of a hardware ramp. The combination of accelerating revenue and narrowing losses proves the advantage is real and not just a result of a tight market.
The moat is currently strengthening as the software stack matures and the fleet of deployed systems grows.
Revenue beat consensus by 7% in Q1 2026 while reaching positive Adjusted EBITDA.
Invested heavily in $390M FCF loss last year to build cloud supercomputers.
Andrew Feldman is a co-founder with a significant stake and proven track record.
Capital Allocation Track Record
Andrew Feldman and his team have demonstrated exceptional strategic judgment by building a commercially viable alternative to the dominant GPU architecture. Their ability to raise massive amounts of capital and secure multi-billion dollar "Sovereign AI" contracts proves they are trustworthy operators who can navigate the most competitive market in technology. They have consistently hit their production and revenue targets while moving the company toward profitability faster than many analysts expected.
The primary governance risk is the company's high dependence on its founders and a concentrated customer base. As a founder-led company, the vision and execution are deeply tied to Feldman, making leadership continuity a key factor to watch. However, they have built a credible bench of experienced executives from the semiconductor and networking industries to manage the massive operational ramp now underway.
We expect revenue to grow from $0.9B in FY2026 to $30.3B in FY2031 (~104% CAGR), with EPS growing from $-0.98 to $26.00. The Wafer-Scale Engine is winning massive sovereign AI and hyperscale contracts as a high-performance alternative to traditional GPU clusters. Manufacturing costs for the massive single-wafer chips drop significantly as production yields improve and fixed R&D is spread over more units. EPS grows faster than revenue because the company is moving from heavy losses to high-margin hardware and software sales. Operating margin expected to reach ~42% by FY2031.
Sovereign AI builds drive massive multi-year revenue backlog. National governments building independent AI hubs provide a stable, long-term source of multi-billion dollar contracts.
Cloud services segment scales to become the primary profit driver. Rented access to Cerebras hardware offers higher margins and stickier revenue than one-off system sales.
Enterprise adoption of CS-3 systems for private data centers. As large companies seek to control their own AI models, demand for dedicated Cerebras racks expands beyond hyperscalers.
Severe customer concentration leads to revenue volatility. If the partnership with G42 or other major sovereign clients stalls, revenue could drop precipitously in any given quarter.
Nvidia's next-generation architecture closes the performance gap. If standard GPU networking improves faster than expected, the performance premium of wafer-scale chips could erode.
Supply chain bottlenecks for specialized wafer-scale manufacturing. Relying on unique, large-format manufacturing processes makes the company vulnerable to production delays that rivals don't face.
Below is our estimate of current and future fair value, with detailed reasoning and assumptions. Fair value is a judgment, not a fact, and other analysts will likely land on different numbers. Use it as one data point in your research, and apply your own discretion in any investing decision.
We use an EV/Revenue approach (Enterprise Value to Revenue) with a margin bridge to justify the long-term profitability path. This framework fits Cerebras because the company is currently inflecting from GAAP losses to massive scale; revenue multiples best capture the value of the $24.6 billion backlog that earnings multiples would currently miss.
Next year's (FY2027) revenue estimate of $2.67 billion multiplied by a 35x multiple, adjusted for $1.39 billion in net cash, gives a per-share fair value of $425. A 35x multiple sits at the top of the AI hardware peer range (ARM 35x, NVDA 23x, AMD 12x) and is justified by Cerebras' significantly higher YoY growth rate of 94% and its unique wafer-scale competitive moat. We used the consensus FY2027 revenue of $2,670M as the base, which is consistent with the massive deployment schedule required to service the OpenAI partnership.
Cross-checked with a peer-anchored Forward P/E using FY2028 estimates, we arrive at a fair value of $414—within 3% of our primary $425 figure. We applied a 90x multiple to the FY2028 EPS estimate of $4.60; while 90x is high for mature semis, it is appropriate for a company growing earnings at a triple-digit CAGR during its first three years of profitability. This close agreement between the revenue-based and earnings-based approaches confirms that the market is likely underestimating the sheer scale of the earnings inflection coming in 2027 and 2028.
We are assuming that Cerebras can recognize roughly $1.85 billion in annual revenue from its current backlog starting in FY2027. This matches the company’s S-1 disclosure that approximately 15% of its $24.6 billion contract value is expected to be recognized over the next 24 months, representing a massive step-up from FY2026 guidance.
We are assuming the business achieves 30% GAAP operating margins by FY2029 as it shifts toward a cloud-service model. While margins are currently compressed due to hardware ramp costs, the shift from one-time wafer sales to recurring cloud-inference services (which grew 178% YoY in Q1) provides a clear path to high-margin software-like profitability.
We are assuming that global data center and power constraints do not delay the deployment of the CS-3 systems. Management has cited "data centers and power" as the primary bottlenecks for growth; our valuation assumes these constraints remain manageable enough to meet the 94% YoY revenue growth trajectory observed in the most recent quarter.
The single biggest risk is the extreme customer concentration with OpenAI and G42, which together account for the vast majority of the revenue base. Any cancellation or renegotiation of the $24.6 billion backlog would collapse the growth thesis, likely compressing the revenue multiple from 35x to 10x and knocking over $300 off the per-share fair value. Watch for any changes in OpenAI's "inference strategy" or shifts toward in-house silicon as an early exit signal.
Bear case ($190): OpenAI or AWS reduces infrastructure commitments by more than 25% due to internal chip development; or Gross margins fail to rebound from the current Q1 dip, staying below 40% through FY2027.
Bull case ($560): Data center power constraints resolve early, allowing more than 20% of the backlog to be recognized in 2027; or A third major "hyperscaler" contract is announced, diversifying the revenue base away from OpenAI.
Clearthesis wrote this report from 38 sources, including SEC filings, industry research, and recent news.
How did you like this thesis?
Your feedback helps us make reports better for you
© 2026 Clearthesis.ai · Report generated on July 1, 2026
This is an AI-generated analysis for informational purposes only and does not constitute financial advice. Data and analysis may not reflect recent developments if viewed significantly after the generation date. Always conduct your own due diligence before making any investment decisions.