Nvidia Has Been the Undisputed King of the Artificial Intelligence (AI) Revolution. Has the Chipmaker Finally Met Its Match?


There’s no question that Nvidia (NASDAQ: NVDA) has been the principal beneficiary of recent advances in artificial intelligence (AI). The company’s graphics processing units (GPUs) quickly became the gold standard for generative AI, capturing a stunning 92% of the data center GPU market, according to market researcher IoT Analytics. Nvidia has parlayed that dominance into five consecutive quarters of triple-digit year-over-year sales and profit growth.

Many competitors have tried to keep up with the company’s relentless pace of innovation, but none have succeeded. Just this year, Nvidia revised its product release cadence from two years to every year, making it even tougher for rivals to compete.

Start Your Mornings Smarter! Wake up with Breakfast news in your inbox every market day. Sign Up For Free »

However, a recent entrant into the AI marketplace is making waves and could mark the first real competition Nvidia has faced.

A person pushing a virtual AI button, surrounded by various technology icons.
Image source: Getty Images.

Cerebras Systems is an AI company founded in 2016, and there have recently been rumblings of an IPO on the horizon. The company believes that “AI is the most transformative technology of our generation.

Cerebras developed the Wafer-Scale Engine (WSE) — a giant semiconductor that is taking a different approach to accelerating AI. The WSE boasts 4 trillion transistors and integrates 900,000 compute cores and 44 gigabytes of Static Random Access Memory (SRAM) into the chip itself.

Cerebras claims that its unique construction reduces latency — or the lag resulting from data transmission — making the third-generation WSE “the world’s fastest commercially available AI training and inference solution.” In August, Cerebras launched what it called “the world’s fastest AI inference,” which it claims is 20 times faster than Nvidia’s GPU-based solutions at a fraction of the cost.

In a press release that dropped last week, Cerebras updated its claims, saying it tripled its “industry-leading inference performance, setting [a] new all-time record.” The company said its tests with Llama 3.2 — the recently upgraded generative AI model from Meta Platforms — were “16x faster than any known GPU solution, and 68x faster than hyperscale clouds.”

While AI-centric efforts by Nvidia and Cerebras have some overlap, it’s important to take a step back and put the rivalry in context.

Nvidia’s chips have a track record dating back 25 years and have stood the test of time. These GPUs dominate a variety of tasks and markets, including video game graphics cards, data centers, earlier branches of AI, and — most recently — generative AI.



Source link

About The Author

Scroll to Top