Nvidia’s AI dominance faces growing competition from rivals

Nvidia’s technological superiority and skyrocketing revenue show no signs of slowing down.

At the same time, the company’s centralization of power is creating new pressures as capital investment explodes and technological change disrupts the sector.

Nvidia’s graphics processing units (GPUs) aren’t cheap. And as customers look to reduce their dependence on them, several companies are emerging as rivals.

Attention to AI is also evolving. GPUs do most of the training, but inference, or running the AI ​​model to perform tasks, is continuous and cost-sensitive. A wave of startups are developing inference chips that they position as cheaper and more efficient than GPUs.

AI hardware chain companies are often both competitors and partners. Silicon giant Broadcom, for example, designs chips that compete with Nvidia and also provides networking technology to connect its GPUs.

As a result, the playing field is rapidly expanding and becoming increasingly complex, rather than a head-to-head competition, even though NVIDIA still holds a miles lead.

These are the biggest challengers to Nvidia’s dominance.

1. From customers to competitors

Google has been working on Tensor Processing Units (TPUs) for about a decade and has become one of its most formidable competitors.

TPUs have been primarily limited to Google’s cloud and internal workloads. In February, the search giant signed a deal to rent them to Meta. Google is also partnering with cloud company Fluidstack to lease TPUs, another change that positions Google more squarely against Nvidia.

Amazon also designs chips called Trainium for training and Inferentia for inference as low-cost alternatives to Nvidia.

Microsoft and Meta are early in the process. Meta said Wednesday it is developing four new silicon generations over the next two years, and Microsoft recently announced an AI inference chip called the Maia 200.

2. Chip startups are catching the inference wave

Investors are pouring billions into chip startups that are catching the specter wave.

Nvidia similarly wants to come in, paying $20 billion to license technology and hire top talent from Groq, a company founded by former TPU engineers and considered one of the biggest challengers to inference.

Several unicorns have been born in this field. Many have been around since before ChatGPT, but are now thriving as infrastructure spending skyrockets and demand materializes.

Founded in 2015 and valued at $23 billion, Cerebras makes dinner plate-sized “wafer-scale” chips for training and inference, and signed a $10 billion deal with OpenAI in January.

SambaNova, which raised $350 million after acquisition talks with Intel fell through, builds AI hardware and software systems for enterprise customers. (Intel told Business Insider that it plans a multi-year partnership with SambaNova and has invested in Series E.)

And Tenstorrent, last valued at $2 billion, also offers an alternative to GPUs.

3. China factor

China remains Nvidia’s biggest geopolitical headache. The United States is tightening export controls on AI chips, with US regulators claiming that some Chinese labs are training models on restricted hardware anyway, Reuters reported.

Nvidia CEO Jensen Huang has repeatedly warned that blocking sales to China would only accelerate progress there.

Huawei is at the center of these efforts. The nearly 40-year-old telecommunications giant is seen as the closest competitor to Nvidia, which builds chips, servers and networking equipment and operates its own cloud.

Chinese chip startups like Cambricon are also emerging as an alternative to Nvidia

Other competitors include Alibaba and Baidu (China’s equivalent of Amazon and Google), which are designing chips for their respective cloud businesses.

4. Old guard

While deep-pocketed chip companies like AMD, Intel, and Broadcom are vying for some of Nvidia’s AI dominance, Nvidia is moving into their turf as well.

AMD, which has developed GPU competitors and whose CEO Lisa Su is a distant cousin of Huang, has secured deals with major cloud and business customers, including Meta.

Intel, on the other hand, has a strong footprint among large enterprise customers, while Broadcom specializes in networking and custom chips. So even if Nvidia continues to lead in GPUs, Intel stands to benefit.

Any tips? To contact this reporter via email, please specify the following address: gweiss@businessinsider.com Or send a signal at @geoffweiss.25. Use a personal email address, non-work WiFi network, and non-work device. here‘is our guide Share information securely.