Perplexity CEO Aravind Srinivas isn’t mincing words when it comes to establishing his company’s ties and reverence to the tech behemoth Nvidia.
“We still think Nvidia is peerless and singular,” Srinivas said, asserting that Perplexity’s recent partnership with Cerebras Systems for the development of their Sonar model doesn’t signal a shift away from Nvidia.
“The majority of the models we serve in production are on Nvidia GPUs,” he stated, noting that Cerebras’ chips aren’t yet capable of handling both dense models and sparse MoEs with the same efficiency Nvidia’s processors offer.
Cerebras’ announcement of their pivotal role in underpinning the Sonar model, optimized for Perplexity search, seemed to have caused some stir.
Their AI inference infrastructure reportedly enabled Perplexity to achieve speeds of up to 1,200 tokens per second, a feat that led to speculation about a potential dent in Nvidia’s market stronghold.
But Srinivas countered, suggesting this perception was unfounded and fueled by clickbait narratives.
It’s undeniable that Nvidia’s GPUs have set a benchmark in AI applications across various sectors. Their CUDA platform revolutionized parallel computing, reinforcing the company’s formidable reputation.
The Search Market: Meanwhile, Perplexity’s launch of the Sonar model in collaboration with Cerebras marks notable progress in the search technology domain.
Perplexity and others like Sam Altman’s OpenAI are trying to chip away the extremely-lucrative Search market away from Google.
The 10 blue links of Google are threatened by a quick answer to a user’s query from the AI model.
When Perplexity sought to disrupt the Search market, Google answered with a similar product — its AI Search Overviews, as a beta feature in the search engine.
The Nvidia Story: It’s without a doubt that Nvidia has emerged the biggest winner in the AI rush following ChatGPT’s public launch.
The stock has gained nearly a thousand percent since but it now faces an alternative rising alternative.
DeepSeek’s R1 model that beat benchmarks from the likes of OpenAI’s o1 while being supposedly trained at a fraction of the costs rattled the market.
Nvidia fell the most as market worried the moat of the Jensen Huang-led company is giving way.
Amid this, US tech leaders have sought to assure that it isn’t the case and that training foundational models still require immense amount of GPUs.
Meanwhile, concerns have also been raised about the numbers shared by DeepSeek on how much it actually costed to train the model and the methodology used to do so.
Disclamer: Dzambhala Founder owns shares of Nvidia.
Planning your financial journey can be daunting but it doesn't have to be. Fire Fast by Dzambhala helps you understand and plan effectively.
Join the vibrant privacy-ensured Dzambhala community on
Want to give feedback on this story? Write to us.