Cerebras Systems stands on the verge of one of the largest tech debuts in years. The Sunnyvale, California company, known for its massive wafer-scale chips built to power artificial intelligence workloads, filed updated paperwork this month. It now eyes a valuation near $30 billion or higher after multiple price range hikes. Orders have poured in. Some reports peg demand at more than 20 times the shares offered. Yet questions linger. Can a firm still unprofitable at scale convince public investors its unusual approach beats the established order?
Andrew Feldman, co-founder and chief executive, has steered the company through multiple funding rounds and one withdrawn IPO attempt last year. He isn’t selling shares in this offering. His post-IPO stake could top $1.5 billion at current talk of $150 to $160 per share. Feldman built Cerebras around a simple bet. Standard graphics processors from Nvidia work well enough for many tasks. But certain AI jobs, especially inference where models generate responses in real time, demand something different. Something bigger.
The Information first framed the listing as a key test for investor appetite in specialized AI hardware makers. That view holds. Cerebras doesn’t chase the same market as everyone else. Its Wafer-Scale Engine measures 58 times larger than Nvidia’s latest B200 chip. It delivers 2,625 times more memory bandwidth. The entire chip is one giant piece of silicon. No separate dies. No interconnect bottlenecks. The design forces a rethink of how computers tackle massive neural networks.
Financials tell a tale of rapid expansion mixed with heavy spending. Revenue hit $510 million in 2025. That marked 76 percent growth from the year before. A master agreement with OpenAI, valued above $20 billion, promises 750 megawatts of inference capacity. It can expand to two gigawatts. AWS signed on too. The cloud giant plans to fold Cerebras CS-3 systems into its Bedrock service. These deals give visibility. They also highlight concentration risk. One customer once made up the bulk of revenue. That picture has improved. Still, public market analysts will watch every quarterly update.
But the company lost money for years. It posted a net loss of $485 million in 2024 before swinging to an $88 million profit in 2025. Gross margins face pressure. Newer systems cost more to build. Competition never sleeps. Nvidia keeps releasing faster GPUs. Startups and big incumbents alike chase custom silicon. Cerebras must prove its hardware delivers enough performance per dollar, per watt, to justify the switch.
Recent market signals look strong. Bloomberg reported the price talk jumped from $115-$125 to $125-$135, then again toward $150-$160. Share count rose to 30 million. A deal at the top could pull in nearly $5 billion. Indications of interest exceeded $10 billion before formal marketing even started. That’s rare heat. It echoes the frenzy around other AI names but centers on actual silicon, not just models.
Feldman has said the shift from training to inference changes everything. Training needs raw power for weeks at a time. Inference runs constantly, serving millions of users. Memory bandwidth and low latency matter more. Cerebras claims its architecture shines here. One system replaces dozens of GPU racks. Data centers save on space, power, cooling. Those claims will face real-world scrutiny once the company reports as a public entity.
Investors have bid up anything tied to AI infrastructure. CoreWeave went public last year. Several chip-related names followed. Yet many AI hardware hopefuls remain private or stumbled. The stakes feel higher now. Power consumption for data centers has become a national topic. Governments worry about grid capacity. Companies hunt efficiency gains wherever they exist. If Cerebras can deliver on its promises, it carves a lasting niche.
The original confidential filing came in 2024. A national security review tied to an Abu Dhabi investor slowed things. The company withdrew plans in October. It raised $1 billion earlier this year at a $23 billion valuation. Backers include Tiger Global, AMD, and Fidelity. That round gave breathing room. The refreshed S-1, filed in April and amended in May, shows remaining performance obligations of $24.6 billion. That’s backlog any hardware firm would envy.
So what makes this listing different? Size, for one. A multibillion-dollar raise at a $30 billion-plus valuation would rank among the biggest tech IPOs ever. It also arrives at a moment when public markets test whether AI spending can sustain high multiples. Nvidia trades at enormous multiples because its GPUs dominate training. Cerebras bets inference will become the bigger long-term market. Analysts at Futurum Group called the S-1 a potential signal that GPU uniformity may fade. Customers want options. They want specialized silicon that matches their exact workload.
Of course risks abound. Manufacturing the world’s largest chip isn’t simple. TSMC produces the wafers. Yields, costs, and scaling all matter. Any hiccup shows up in margins. The company acknowledges near-term pressure there. Then there’s the broader economic picture. Interest rates, trade tensions, and energy prices could sway buyer budgets. AI hype has cooled in spots. Enterprises now ask for clear return on investment before signing big checks.
Still, the order book tells its own story. Multiple upward revisions in days. Strong institutional interest. Underwriters Morgan Stanley, Citigroup, Barclays, and UBS have their hands full. Pricing is set for May 13. Shares will trade under ticker CBRS. Early trading could swing wildly. Many recent hot listings popped then settled. Long-term performance will depend on execution.
Cerebras isn’t the only name testing these waters. Other AI chip firms watch closely. So do pure software players pondering their own listings. A successful debut would open doors. It would validate hardware innovation outside the dominant supplier. A weak showing might tighten capital for the whole category. The market has spoken loudly in pre-IPO demand. Now comes the harder part. Delivering numbers that match the narrative.
Feldman and his team spent years refining the wafer-scale idea. They bet against conventional wisdom that chips must get smaller. They built software to program the giant processor efficiently. They lined up cloud partnerships so customers don’t need to buy hardware outright. That combination, massive scale plus flexible delivery, could prove powerful. Or it could become an expensive experiment if adoption lags.
Recent X chatter shows excitement. Traders note the repeated price hikes as proof of scarcity in AI compute. One post called it a major demand signal for anything tied to solving power and performance bottlenecks. Sentiment aligns with the banking book. Whether that carries through earnings season is another matter.
In the end, Cerebras offers public investors a direct bet on a different way to run AI. Not faster GPUs. Not just more of them. A fundamental redesign of the processor itself. The company has the contracts. It has the technology. It now needs to show it can turn that into consistent profit while fending off competitors racing to catch up. The listing won’t settle the AI chip wars. But it will set a price on one vision of the future. And Wall Street is paying attention.
WebProNews is an iEntry Publication