Cerebrra’s become the world’s fastest host for Deepseek R1, who exceeds Nvidia GPUs with 57x

Cerebras systems Announced today that it will organize the breakthrough of Deepseek R1 Artificial intelligence model on American serversPromising speeds up to 57 times faster than GPU -based solutions, while sensitive data is stored within American boundaries. The move comes in the midst of growing concerns about the fast AI preface and data privacy of China.

The AI ​​Chip Startup will implement a 70 billion parameter version of Deepseek-R1 Run on its own hardware on wafer scale, which provides 1,600 tokens per second-a dramatic improvement compared to traditional GPU implementations that have struggled with newer “reasoning” AI models.

Why reform the Reasoning models of Deepseek de Enterprise AI

“These reasoning models have an influence on the economy,” said James Wang, a senior executive at Cerebrra’s, in an exclusive interview with Venturebeat. “In principle, every knowledge worker must perform a kind of cognitive tasks with multiple steps. And these reasoning models will be the tools that introduce their workflow. “

The announcement follows a tumultuous week in which the rise of Deepseek has caused Nvidia The biggest loss of market value ever, almost $ 600 billionCalling questions about the AI ​​supremation of the chip giant. Cerebras’s solution is immediately engaged in tackling two important concerns that have emerged: the calculation requirements of advanced AI models and data display.

‘If you use it Deepseek’s APIWhat is now very popular that data is sent directly to China, “Wang explained. “That is a serious reservation that [makes] Many American companies and companies … not willing to consider [it]. “

How Cerebras’s wafer scaling technology beats the traditional GPUs with AI speed

Cerebras reaches its speed benefit due to a new chip architecture that holds entire AI models in a single wafer format processor, which eliminates the based on memory based on GPU-based systems. The company claims the implementation of Deepseek-R1 competitions or surpasses the performance of OpenAi’s own models, while it runs entirely on American soil.

The development represents an important shift in the AI ​​landscape. DeepFounded by former Hedgefonds director Liang Wenfeng, the industry shocked by advanced AI reasoning possibilities to only only 1% of the costs of American competitors. Cerebras’s hosting solution now offers American companies a way to make this progress while retaining data control.

“It is actually a nice story that the US Research Labs have given this gift to the world. The Chinese took it and improved it, but it has limitations because it is in China, has some censorship problems, and now we take it back and perform it on American data centers, without censorship, without data retention, “Wang said.

US Tech Leadership is confronted with new questions, because AI innovation goes worldwide

The service is available through a Developer Preview From today. Although it will initially be free, Cerebras is planning to implement API Access Controls Because of a strong early question.

The move comes as the American legislators struggle with the implications of Deepseek’s rise, which has exposed potential restrictions in American commercial restrictions Designed to maintain technological benefits compared to China. The ability of Chinese companies to achieve breakthrough AI possibilities, despite Chip -Export checks has requested calls for new legal approaches.

Industrial analysts suggest that this development could accelerate the shift of GPU-dependent AI infrastructure. “Nvidia is no longer the leader in the conclusion,” Wang noted, pointing to benchmarks that show superior performance of various specialized AI chips. “These other AI chip companies are really faster than GPUs for performing these latest models.”

The impact extends beyond technical statistics. Because AI models are increasingly taking advanced reasoning opportunities, their computational requirements have been raised. Cerebras claims that his architecture is better suited for this emerging workload, which may reform the competitive landscape in the implementation of Enterprise AI.