How Cerebras Is Betting Big on a Different Kind of AI Chip

Cerebras Systems just secured $850 million in borrowed money to fuel growth before taking the company public. The funding comes as the company tries to convince the world that its AI processor — called the WSE-3 — can outperform the chips that currently dominate the market, which are made by NVIDIA.

Cerebras Systems announced the credit line without saying which banks are involved. The company's chief financial officer, Bob Komin, said the money will support operations and expansion.

The timing matters because Cerebras plans to go public soon. In July 2024, the company quietly filed paperwork with the Securities and Exchange Commission to start the IPO process. This credit facility gives the company money to spend while preparing to sell shares to the public — a long and expensive process.

What Makes the WSE-3 Different

The WSE-3 is Cerebras's main product, and it works in a fundamentally different way than the chips most AI companies use today.

Think of it this way. Most AI systems use multiple GPUs — imagine a team of workers collaborating on a large project. They sit in different locations, talk to each other over fast cables, and coordinate their work. It works well, but communication between them takes time and requires extra infrastructure.

Cerebras takes a different approach. Instead of using multiple separate chips, it puts everything onto one giant wafer — the underlying piece of silicon. All the computing power and memory sit together on a single piece, so there is no delay waiting for data to travel between chips.

Cerebras says its WSE-3 is 58 times larger than the biggest NVIDIA GPU available. The company claims this size gives it an edge, especially for training and running large language models — the AI systems behind tools like ChatGPT.

Why the Big Loan Now

For a company making complex hardware, an $850 million credit line is a serious commitment. Banks only lend this much money to companies they believe will succeed.

This type of loan works differently than selling stock. Instead of giving up a piece of the company to get money right away, Cerebras can borrow as needed — like a corporate credit card. This lets the company keep more control and ownership before going public.

Hardware companies need this kind of financial cushion. Building AI chips requires expensive factories, long manufacturing timelines, and money to pay suppliers upfront. The specialized equipment and processes Cerebras uses add even more to that cost.

The Bigger Picture

Right now, NVIDIA dominates the AI chip market. The company's GPUs power most large AI systems because they work well, have mature software, and are what developers know how to use.

But Cerebras is not the only company trying a different approach. Several other startups are also building alternative chips they say perform better or more efficiently than GPUs.

Here is what is worth considering. We have seen architectural competitions like this before in technology history. When graphics processors first emerged as an alternative to traditional central processors, the winner was not just the best performing technology — it was the one that built the largest community of developers and the most complete software ecosystem. Cerebras faces that same challenge now. It must prove not only that its chip works well in labs but that real companies will adopt it for their actual AI systems.

The Real Test Ahead

On paper, Cerebras's design has advantages. By eliminating delays between chips, the company can potentially process certain kinds of AI tasks faster.

But there are real obstacles. The first is manufacturing. Making a perfect wafer with no defects is extremely difficult. Smaller chips can be tested in pieces; Cerebras must make the entire wafer work reliably.

The second is integration. Data centers have built their infrastructure around NVIDIA's chips. They know how to cool them, power them, and connect them. Cerebras chips may need custom setups, which means more work and expense for customers.

The third is software. GPU technology has decades of maturity behind it. Engineers have spent years building software that takes advantage of how GPUs work. Cerebras software is newer and less battle-tested.

What Comes Next

The credit facility gives Cerebras the financial breathing room to scale up production and work toward its IPO. But money alone will not determine success.

The real question is whether the company can convert its technical advantages into customers who actually buy and use its chips. That means proving the WSE-3 solves real problems for real businesses — not just winning benchmarks in the lab. The companies that win in AI hardware will be those that combine good technology with solid software support, reliable manufacturing, and deep relationships with customers. For Cerebras, all of that work still lies ahead.