Decentralized AI Network on Solana Faces Challenges

According to PANews, a decentralized computing network built on Solana, known as io.net, has raised funding from Mult1C0in Capital and Moonhill Capital. The exact amount of funding has not been disclosed. The platform, which falls under the Depin and AI sectors, provides instant, permissionless access to global GPU and CPU networks for machine learning training on GPUs. It boasts 25,000 nodes and uses revolutionary technology to cluster GPU clouds together, saving up to 90% of computing costs for large-scale AI startups. However, the platform has been scrutinized for the number of GPUs it claims to have. Four different figures have been suggested: 7,648 (attempted during deployment), 11,107 (manually calculated from their resource manager), 69,415 (an unexplained, constant number), and 564,306 (a number without any supporting, transparent, or substantial information). The actual number of GPUs is believed to be 320. This figure is based on the fact that all GPUs are listed as 'free' on the resource manager page, but none can be rented. The only GPUs that can actually be rented are 320. If they cannot be rented, they are not considered to exist. The decentralized AI protocol has been criticized for several issues. Firstly, there is no cost-effective and time-efficient way to conduct useful online training on highly distributed general hardware architectures. This would require a significant breakthrough. Secondly, inference on general hardware sounds like a good use case, but the rapid development of hardware and software means that a general decentralized approach performs poorly in most key use cases. Thirdly, inferring from correctly routed requests, coexisting with the GPU cluster, and using decentralized cryptocurrency to lower capital costs to compete with AWS and incentivize enthusiasts to participate sounds like a good idea. However, due to the many suppliers and the scattered liquidity of the GPU spot market, no one has integrated enough supply to provide for those operating real businesses. Lastly, the software routing algorithm must be very good, otherwise there are many problems in the operation of the general hardware of the consumer operator. Unpredictable supply will cause trouble for operations and bring uncertainty to demand-side requesters. These are all tricky problems that will take a long time to solve. All bids are just a joke.