Jensen Huang, the head of NVIDIA, said in an interview with CNBC, “If you want to start a startup today, it’s you and AI. And you’re supercharged by the AI supercomputer.” NVIDIA sees AI and startups in everyone. In turn, for everyone else, NVIDIA and AI are increasingly getting synonymous. So much so, that as soon as we see a new AI startup or hear about a company raising funds, we are sure to stumble upon NVIDIA, either as a backer or a GPU provider! 

Companies buying NVIDIA GPUs through VC funds is one thing, but what if NVIDIA funds startups so that they can buy NVIDIA GPUs? Funnily enough, this is what is happening now.

Inflection AI, a year-old generative AI startup, just raised $1.3 billion in funding from Microsoft and NVIDIA. And guess what? The company is now building the world’s largest AI cluster consisting of 22,000 H100 GPUs from NVIDIA. For comparison, OpenAI’s ChatGPT was reportedly trained on 10,000 A100s, which are actually a predecessor to NVIDIA’s H100s.

At the same time, NVIDIA, along with Google and Salesforce, also announced a funding of $141 million for RunwayML, another generative AI startup. Earlier this month, NVIDIA also invested in another Google-backed OpenAI rival, Cohere. The $270 million funding includes Oracle and Salesforce as well. For this, Oracle said that Cohere will offer its AI services on Oracle’s data centres, which run on 16,000 NVIDIA chips. Even CoreWeave has been running on NVIDIA GPUs. 

The list of NVIDIA-backed companies goes on and on. The company that was minting billions of dollars is now worth trillions. It all started with NVIDIA announcing its Inception AI startup program. According to PitchBook, by 2021, NVIDIA represented two-thirds of the total AI startups in the world. 

NVIDIA wants a piece of every startup. Synthesia, the video generation platform, gained unicorn status in June by raising funds from NVIDIA. Obviously, most of that amount would be spent on buying NVIDIA GPUs. But interestingly, after the funding, Synthesia’s founder said, “While we weren’t actively looking for new investment, NVIDIA shares our vision for transforming traditional video production into a digital workflow.” 

GPU’s are the real heroes in AI and NVIDIA knows it

To put it in simple words, NVIDIA’s strategy is – “We give you money, you buy our GPUs, and increase your valuation”. A common joke doing the rounds is that “technology companies may be hiring more GPUs than employees”. For the companies, NVIDIA’s backing and branding have been boding well and helping them make their millions. 

This strategy has been working in favour of NVIDIA when it comes to competing with rivals like AMD or Intel. NVIDIA holds 88% of GPUs market share in the world, leaving 12% with AMD and Intel. There is no doubt that the company is the biggest benefactor of the generative AI race. 

Interestingly, the company doesn’t have to care about any rivalry between Microsoft, Google, Meta, OpenAI, or anyone else, since every single company is using NVIDIA GPUs to run their models. Companies having the largest cluster of NVIDIA GPUs even boast about how many GPUs they use. It’s almost akin to boasting about how much money and valuation one has since one NVIDIA GPU for building AI models costs $10,000.

Elon Musk, the head of Tesla and Twitter, also emphasised the importance of GPUs in April. He prophesied that everyone would be lining up to buy GPUs, which is clearly the case now. Musk is also expected to soon enter space with its Dojo Supercomputer. Though for this, Musk has said that he would not sell the GPUs but just provide Tesla’s supercomputer as a service, not competing with NVIDIA.

NVIDIA has a chance to pull a prank on everyone

It does seem like Huang took Warren Buffet’s famous quote, “The best investment you can make is an investment in yourself,” pretty seriously. And now, he has a chance to pull the biggest prank on everyone — instead of giving more money to startups to buy its GPUs, the company can call back all its GPUs and build its own model. Or maybe just give away its GPUs for free. You can do it NVIDIA!