The Wafer Giant Threatening NVIDIA's GPU Hegemony

Summary

• The Blockbuster Debut: Cerebras went public on May 14, 2026, pricing at $185 per share (well above expectations) and raising $5.55 billion in a heavily oversubscribed IPO.
• The Hardware Revolution: Its flagship Wafer-Scale Engine 3 (WSE-3) is a single, massive silicon wafer that destroys the traditional "memory wall" throttling standard GPU clusters.
• Monstrous Backlog: Backed by a historic $20 billion OpenAI deal and an innovative AWS partnership, Cerebras boasts an unparalleled $24.6 billion forward backlog.
• The Premium Catch: While the tech is undeniably superior, the stock trades at an astronomical trailing multiple, faces intense customer concentration, and its current GAAP profitability relies on an accounting anomaly.

Company Overview: Subverting Chip Orthodoxy

For decades, the semiconductor playbook has been simple: etch chips onto a silicon wafer, chop them up into individual dies, and connect them using external cables. The problem is that small defects can ruin a wafer, so chips are kept small to protect manufacturing yields.

Cerebras flipped this orthodoxy on its head. Its Wafer-Scale Engine 3 (WSE-3) is a single, giant, uncut piece of silicon. Measuring 46,225 square millimeters (roughly 58 times the size of a flagship GPU), it packs 4 trillion transistors and 900,000 AI cores. Cerebras bypassed the yield issue by building microscopic redundancy directly into the design. If a core fails, the system automatically routes workloads around it.

By keeping the wafer intact, Cerebras keeps 44 gigabytes of lightning-fast SRAM memory mere nanometers away from the compute cores. This eliminates the severe latency penalties that occur when traditional GPUs have to retrieve data from external memory pools.

Macrotheme: The Ascendancy of Inference and Agentic AI

To understand why the market is paying a massive premium for Cerebras, we have to look at how AI workloads are evolving. The industry is moving rapidly from the training phase (building models) to the inference phase (running models). According to McKinsey data, AI inference workloads are projected to grow at a 35% CAGR through 2030, completely overtaking training.

This hardware shift is being turbocharged by the rise of Agentic AI. Early AI chatbots were built for human reading speeds (20 to 50 tokens per second). Agentic workflows, however, use autonomous AI agents that write code, debug software, and call APIs completely behind the scenes.

In an agentic loop, bots talk to other bots. Because these loops are highly iterative, latency compounds aggressively. If a 50-step reasoning loop takes minutes on standard GPUs, the workflow breaks down.

By pushing raw speeds past 2,100 tokens per second on open-source models like Llama 3.1, Cerebras compresses these complex machine loops from minutes into fractions of a second.

Bullish Catalysts

1. A Staggering Multi-Trillion Dollar Addressable Market

Cerebras is positioned at the absolute center of a rapidly expanding market. The global shift from training models to running active, live inference workloads is unleashing an unprecedented wave of infrastructure spending. Bloomberg Intelligence forecasts that the generative AI market will expand to $2.3 trillion by 2032. Within that ecosystem, the hardware inference market alone is projected to grow at a blistering 32% compound annual rate to hit $1.3 trillion. Because Cerebras designed an architecture specifically optimized to handle this exact tsunami of inference traffic, its total addressable market is expanding exponentially.

2. The $20 Billion OpenAI Mega-Deal

In late 2025, Cerebras signed a transformative Master Relationship Agreement with OpenAI. OpenAI is contractually bound to procure 750 megawatts of inference capacity through 2028, backed by a $1.0 billion working capital loan to fund Cerebras's manufacturing scale-up. The first major fruit of this union is GPT-5.3-Codex-Spark, a model explicitly optimized for Cerebras hardware that cranks out real-time code at over 1,000 tokens per second.

3. The AWS "Disaggregated Inference" Partnership

In March 2026, AWS integrated Cerebras CS-3 systems directly into Amazon Bedrock. This partnership solves a classic engineering headache by splitting the inference workload into two parts: AWS Trainium3 chips handle the compute-heavy "prefill" phase, while the Cerebras wafer dedicates its massive SRAM bandwidth entirely to the sequential "decode" phase. The result is up to a 15x speed improvement over monolithic GPU setups.

4. Absolute Benchmark Dominance

The hardware advantages translate directly into undeniable, third-party validated dominance. Data from independent benchmarking firm Artificial Analysis shows that Cerebras has systematically dismantled existing inference latency records across leading open-source models:

To put that in perspective, a massive, highly intelligent 70-billion-parameter model running on Cerebras hardware operates substantially faster than a tiny, stripped-down 3-billion-parameter model running on state-of-the-art competitor GPUs. For enterprise developers, this means they can finally deploy hyper-intelligent models without sacrificing user experience or hitting a latency wall.

5. Clear Backlog Visibility

Cerebras entered the public markets with a staggering $24.6 billion in Remaining Performance Obligations (RPO). Management's revenue recognition schedule locks in an annualized forward revenue baseline of roughly $1.845 billion through 2027 purely from drawing down this backlog. This provides an incredibly stable revenue floor in a highly volatile semiconductor market.

Risks

1. Supply Chain Fragility

Cerebras has zero long-term, committed manufacturing capacity allocations from TSMC. Every wafer is bought via ad-hoc purchase orders, leaving them highly vulnerable if TSMC capacity tightens.

2. Dilution Overhang

Additionally, public retail shareholders face structural subordination. Insiders hold Class B shares with 20-to-1 super-voting power. Even worse, the company issued massive equity warrants to OpenAI (33.4 million shares at a near-zero strike price) and AWS (2.7 million shares) to secure their partnerships. As these vest, strict accounting rules will force Cerebras to log them as contra-revenue, which will artificially suppress top-line growth metrics in 2026 and 2027.

3. High Customer Concentration

Cerebras famously had to pull its initial IPO filing due to CFIUS regulatory concerns regarding its extreme revenue reliance on UAE-based G42. While they successfully restructured that relationship, they essentially traded sovereign concentration for corporate concentration. OpenAI dominates the forward backlog. If OpenAI alters its architecture or shifts its infrastructure strategy, Cerebras's primary engine of future growth vanishes.

4. The Accounting Mirage and Cash Burn

At first glance, Cerebras boasts an impressive 2025 GAAP net income of $237.8 million on $510 million in revenue. However, a forensic look at the ledger reveals that this net income includes a one-time, non-cash $363.3 million paper gain from restructuring old Middle Eastern contracts. Once you strip away this accounting adjustment and stock-based compensation, Cerebras actually posted a non-GAAP operating loss of $75.7 million for 2025. Furthermore, cloud service margins compressed from 68% to 30% over the year due to massive upfront capital expenditures required to build out data centers for OpenAI.

5. The Mid-November Supply Shock: A Staged Lock-Up Avalanche

Unlike a traditional IPO where insider shares are locked down tightly for a clean 180 days, Cerebras engineered a complex, nine-tranche release schedule. Non-executive employees have already been allowed to cash out small slices of their holdings during the first few days of trading, with subsequent chunks tied directly to quarterly earnings reports throughout 2026.

The real test arrives in mid-November 2026. Two trading days after the company reports its Q3 2026 earnings, the final and largest phase of the lock-up expires. This milestone releases the shares held by directors, corporate officers, and major pre-IPO investors, suddenly hitting the market with approximately 171.1 million newly eligible shares.

For retail investors, this creates a massive near-term supply risk. Many early venture capitalists backed Cerebras at valuations as low as $8 billion. At the current premium stock price, these insiders are sitting on astronomical gains. Even if the operational momentum remains strong, the temptation to lock in profits could create a heavy wave of secondary selling pressure. While premier early backers like Benchmark and Fabrica have publicly stated they have no immediate plans to sell, the sheer volume of newly unlocked shares could easily trigger severe volatility for such a highly valued name.

Valuation: Priced for Absolute Perfection

Evaluating Cerebras Systems requires us to bridge a massive gap. If you look at standard trailing metrics, the valuation borders on the absurd. At a market cap of over $50b, the stock trades at well over 100 times its historical revenue.

However, the market is not pricing Cerebras on what it did last year. It is pricing the company on its unprecedented $24.6 billion forward order book.

To find a fair value, we have to look five years into the future using a customised growth model. Based on management's delivery schedule for the OpenAI tranches and the aggressive rollout of AWS cloud capacity, a 55% annual revenue growth rate over the next five years is highly achievable. This aggressive top-line expansion will fundamentally change the company's profitability.

Right now, Cerebras is burning cash to physically build out the data centers required to host its giant chips. Within five years, that heavy capital expenditure will stabilise. The company will transition from a hardware builder into a cloud service provider, and we can expect profit margins to expand significantly to a target of 25%.

Finally, applying a Future PE of 35x brings the valuation back down to earth, aligning it with other elite, dominant tech monopolies once their hyper-growth phase matures.

When you plug these estimates into the valuation model, the current premium begins to make sense. You are paying a steep upfront price today for guaranteed, contracted cash flows tomorrow.

If Cerebras executes its delivery pipeline flawlessly without supply chain hiccups, the current share price will look like a bargain in hindsight. But make no mistake, any delays in manufacturing will cause this high-flying valuation to compress rapidly.

The Bottom Line

Cerebras is a classic high-conviction, high-beta AI infrastructure asset. The company correctly anticipated that memory bandwidth and latency would become the ultimate physical bottlenecks of the agentic AI era. While the stock commands an undeniable premium and carries genuine operational friction, its technological moat and multi-year backlog make it one of the most compelling infrastructure bets on real-time AI execution.

The long term opportunity here is simply too big to ignore, but it's critical to be highly tactical about an entry point. Between the astronomical current valuation, messy underlying margins, and the massive insider share unlock looming in mid-November, the stock is bound to face severe near term turbulence. Because of these factors, I am lowering my price expectations for the immediate future. Rather than chasing the hype at today's hefty premium, I'm planning to dollar cost average into the name, building a position steadily on major pullbacks to safely capture the massive long term upside.

Not financial advice. DYOR.