Cerebras just announced 6 new AI datacenters that process 40M tokens per second — and it could be bad news for Nvidia

Cerebras systemsA AI hardware start that is constantly difficult Nvidia’s preference The artificial intelligence market announced a significant expansion of two major enterprises, which allows you to become a leading provider of the data center footprint and the company’s high-speed AI inferences and the company’s high-speed AI.

The company will add six new AI data centers across North America and Europe, increases the power of more than 40 million token per second. Expansion, Dallas, Minneapolis, Oklahoma City, Montreal, New York and France, include 85% of the total capacity of the United States.

“This year, our goal is to all our needs and all the new requirements, as a result of new models, Llama 4 and new Deepseek models,” James Wang, VentureBeat in an interview with product marketing James Wang. “This year this year is our great growth initiative this year.

The expansion of the data center represents the company’s ambitious bet that the market of high-speed AI inferential market – experienced AI models create speeches for real world applications – companies are looking for faster alternatives to GPU-based solutions from NVIDIA.

Strategic partnership that brings high speed AI developers and financial analysts

Along with the expansion of infrastructure, Cerebras announced partnership Hug facePopular AI Developer Platform and In consequenceA market intelligence platform widely used in the financial services industry.

This Hug face Integration will allow five million developers to enter Cerebras account With one click without signing up for cerebras individually. This represents a basic distribution channel for developers working with source models, especially Call 3.3 70B.

“Hugging the face is AI and all open source AI development center” Wang explained. “Integration is super beautiful and is not released. You simply appear in the list of their results. You can simply check the box and then use Cerebras.”

The AlphaSense Partnership, with the Financial Intelligence Platform, “Global, the first indoor-source AI model seller AI model seller AI model seller” represents customer earnings with a financial intelligence platform. The company that serves about 85% of Fortune 100 uses cerebres to speed up search opportunities working in AI for market intelligence.

“It’s a very big customer victory and a great deal for us,” Wang said. “We speed them up to 10x, so when you take five seconds or longer, it is mainly converted to cerebras.”

How Cerebras wins the race for the EI result speed, as well as substantial models slow

Cerebras claims itself as a high-speed resulting specialist Waffle-scale engine (WSE-3) processor AI models can work 10 to 70 times faster than GPU-based solutions. This speed advantage has become increasingly valuable to develop AI models towards more complex thinking skills.

“If you listen to Jensen’s speech, the justification is the next big thing, and even according to NVIDIA, Nvidia CEO Jensen Huang said.

This slowing, special equipment creates an opportunity for jerebras designed to speed up more complex AI workloads. The company has already provided high-profile clients including Confusion ai and You have the mistralThose who use Cerebras to strengthen the EU search and assistant products.

“We help the world’s fastest AI search engine in the world. It is not possible in another way,” Wang said. “We help the Mistral achieve the same feat. Now the reasons for people to subscribe to Le Chat Pro, before, your model is probably the same level as GPT-4.”

Cerebras’ difficult and forced economy behind the call to Nvidia

Cerebras also bets that the combination of speed and cost, even the use of leading models such as GPT-4 will be involved in their effects.

Wang pointed out that the methane Call 3.3 70BAn open source model optimized for Cerebras’s hardware, now the same as the exploration tests like Openai’s GPT-4, although it is significantly less.

“Anyone who uses GPT-4 can move only as a drop-down substitution to LLAM 3.3 70B today,” he said. “The price for GPT-4 [about] $ 4.40 in mixed conditions. And Llama is like 3.3 to 60 kopecks. 60 cents, are we right? Almost costs almost a measurement order. And if you use Cerebras, you increase the speed with another size order. “

Serebras’ tornado-proven information centers built for EU continuity

The company prepares solid infrastructure-based investments as part of the expansion. The Oklahoma City mechanism, which is scheduled to come online in June 2025, was designed to face excessive air events.

“Oklahoma, as you know, is a kind of tornado zone. Thus, this information center is actually rated and designed to be fully resistant to tornadoes and seismic activity,” Wang said. “This will endure the most powerful tornado registered so far. This thing is only crossed, this thing will only continue to send the LLA Tokens to developers.”

Oklahoma City Institute working in partnership with measurement Datocenter will be more than 300 cerebras CS-3 systems And tripled unnecessary power plants and special water-cooling solutions, especially for Cerebras’ waffle systems.

From skepticism to market administration: How cerebras prove its value

Today, the expansion and partnerships are announced, represent a significant stage for Cerebras trying to prove itself in an AI device market Nvidia.

“I think what happened to the customer speech is what was reasonable, perhaps when we first started, I think it takes into account the diversity of the logos put into bed now.”

The company targets three specific areas provided by the fastest result: real-time sound and video processing, justification models and coding applications.

“Coding is one of these kinds of justifications between perhaps to a minute to create the entire code, as well as the justification and regular Q & A,” he said. “Speed ​​is directly proportional to the developer productivity. Therefore there is speed there.”

Ceremras, a place where all the AI ​​works among the workloads, found a place where he could require leadership on the largest cloud providers.

“No one is generally fighting against AWS and AZURE on their scale. It is obviously they will not be fully scaled, but to repeat a key segment … We will have more than them to come to a high-speed result,” said Wang.

Why Cerebras’s EU sovereignty and US-centered expansion issues for future work loads

Expansion is a period where the AI ​​industry is focused on the expiration opportunities, because companies are income for the fact that companies have passed since their generation work to apply the production of the manufacturer’s speed and cost efficiency.

85% of the U.S., Cerebras, once the main player in the development of local AI infrastructure, which is converted to the national priority of technological sovereignty.

“Cerebras is turbocharging the future of the AI ​​leadership, which is not comparable to the EU leadership of the United States – this new global information will play the role of a wave of ai innovation,” he said.

Models of thinking like DeepSeek R1 and Openai’s O3 Become more common, the requirements of the solutions for faster results are likely to increase. These models that may take minutes to create a response to traditional apparatus, according to the company, operate near the cerebras systems.

Evaluating the AI ​​infrastructure options for technical decision makers, the expansion of Cerebras, especially the response period, represent a significant new alternative to GPU-based solutions for applications where the user practice is critical.

The company demonstrates a clear strategy to actually read the NVIDIA’s dominance in a wider AI hardware market, but aimed at high-speed result and fundamental infrastructure investments, a rapidly developing AI view.