Artificial intelligence (AI) projects in the cryptocurrency space could require investments in graphics processing units (GPUs) equivalent to their total market capitalisation to achieve their goals. The creator economy benefiting from AI-generated videos will require more GPUs than any major tech company currently has.
The world's largest and most established hub for all things crypto-currencies, blockchain and the Web3 is highlighting the possibility of text-to-video generation, galvanising the crypto-currency market. Interest in AI skyrocketed when OpenAI unveiled a demo of Sora. However, making the technology mainstream would require staggering computing power, requiring more server-grade H100 GPUs than Nvidia produces in a year.
To make text-to-video generation commonplace will require hundreds of thousands of GPUs, more than Microsoft, Meta and Google currently use combined. Emerging crypto AI projects also promise to achieve text to video and image generation, and the AI token category now represents a market capitalisation of $25 billion according to CoinGecko data.
To support the community of creators on TikTok and YouTube, 720,000 high-end Nvidia H100 GPUs will be needed. Sora requires up to 10,500 powerful GPUs to train for a month, and can only generate around 5 minutes of video per hour per GPU for inference. Widespread adoption means that inference will surpass training in terms of computing power usage.
While Nvidia is synonymous with the AI revolution, AMD also has competing products, and other ways exist to outsource computing power to GPU farms. Render (RNDR) and Akash Network (AKT) offer distributed GPU computing, but the majority of GPUs on these networks are gaming-grade and therefore less powerful than Nvidia's server-grade GPUs or those from AMD.
