The GPU shortage will drive mainstream adoption of Web3 infrastructure.
To train artificial intelligence (AI) models, high-end graphics processing units (GPUs) like the NVIDIA A100s and H100s are required. However, these GPUs are only needed for a limited period of time for each model, are very expensive and have a constrained supply – this combination means owning them is out of reach for many startups.
Amy James and Devon James are co-executive directors of the Web3 Working Group.
Further, many people have serious concerns about AI safety, from worries about how deep fake images could harm people or the economy to fears of a Terminator-like malevolent AGI determined to destroy society. [AGI stands for artificial general intelligence, meaning an autonomous system that surpasses human capabilities, aka “the Singularity.]
Decentralized physical infrastructure networks (increasingly known as DePIN), specifically compute and storage networks, offer the solution and will make AI and Web3 go mainstream for two reasons: access and safety.
Despite existing for nearly a decade, use of Web3 has remained limited to a narrow audience of crypto believers. This has befuddled enthusiasts who had expected Web3 to be adopted even faster than the World Wide Web. After all, the original web protocol had to start from nothing while decentralized Web3 protocols have the benefit of the existing web.
But mainstream users have not yet joined decentralized social networks or use protocols like decentralized file storage because they don’t have a strong enough need for them. For most people, despite awareness of Web2 platforms engaging in mass surveillance and manipulation, these services work “just fine” and their powerful network effects lock users in.
The emergence of AI will lead to a breakout moment for Web3 infrastructure because these protocols offer solutions to the GPU shortage and data challenges faced by AI startups. DePIN will transform AI development and, in the process, onboard the first mainstream users to Web3 protocols.
AI startup founders have the technical ability and motivation to push past the friction that has held back mainstream audiences so far, such as having to buy tokens via a clunky user experience, because these founders require access to high quality, expensive, difficult to acquire GPUs in order to succeed.
Protocols like Akash, often described as “Airbnb for GPUs,” provide decentralized marketplaces for GPU owners to rent their resources. These protocols democratize AI innovation by making high-cost resources accessible to smaller players, all while allowing hardware owners to generate passive income.
And while AI startup founders will come to Web3 protocols to get access to the GPUs they need, they will stay because these protocols unlock new features that aren’t available on the Web, allowing Web3 to compete effectively.
File storage protocols like Arweave disrupt the data oligopoly by providing a one-time-payment model for permanent data storage. Using permanent storage for training data makes machine learning open and verifiable, enhancing trust for AI models.
Right now AI products from large companies like OpenAI’s ChatGPT and Google’s Bard are leading the movement, but this is a case where open source versions have a competitive advantage.