Photonics is proving to be a tough nut to crack
It is likely that the growing compute power necessary to train sophisticated AI models such as OpenAI’s ChatGPT will eventually run up against a wall with mainstream chip technologies. This is because GPUs, which are designed for graphics processing, do not have enough computing power to train these models efficiently. As AI technology becomes more sophisticated, it will be important to find alternative chip technologies that can support these types of models.
This suggests that as AI technologies develop, they are becoming more powerful and requiring more resources to train. This power consumption is likely to continue rising as AI becomes more pervasive, so it’s important that wefind ways to curb its usage if we want to avoid harming the environment in the process.
Current prices for AI hardware could be increasing due to a shortage, though Microsoft may face other issues with its implementation. The ChatGPT-like model is estimated to cost over $4 million to train from scratch, showing the potential difficulty in implementing this technology.
A potential issue with photonic chips, however, is that they are not yet able to transmit data as quickly or accurately as traditional processors. Additionally, the interference caused by light waves could ultimately impede computational performance. If these disadvantages can be overcome, however, it is possible that photonic chips could eventually revolutionize AI training methods by providing faster and more reliable signals.
One reason for the muted response from the photonic technology market may be that it has become increasingly difficult for startups to prove their products’ worth. For many companies, the difficulty of acquiring necessary licenses and forging partnerships with giants such as Intel and NTT means that there is little incentive to enter this relatively nascent market.
Investors seem to be giving up hope that photonic chips specifically designed to accelerate AI research will completely solve all of the technology’s flaws. However, while they may not be a silver bullet, they are still an important part of the puzzle.