banner

Its been quite a journey with AI since the ChatGPT announcement in 2022. We’ve seen many startups built around AI, companies pivoting to new AI features and people adopting the technology to build something cool for themselves. 2025 I guess was the year of agents and this year its about orechestration and scaling agents.

Software is being built at tremendous scale and the number of GitHub PR’s is a good metric to visualize ( X tweet ). However, many experts also feel that its becoming saturated. The scope for a new SaaS or software solution to start from 0 to 1 and hit scale is getting smaller now. Maybe for a few sectors its still green pastures but AI-assisted coding has made it easier for enterprises to proabably build their own software instead of buying it. Take YC class of 2026, majority of them are “AI” startups and doing a simple search shows the sheer number of software AI startups. I have a newsletter that lands in my inbox ever week which also has a section showing the latest AI startups and its basically just AI software startups trying boost productivity, automate, vibe code etc. I’m not sure how each of these startups will get customers when everbody is launching the same thing with a different brand.

I was in Vivatech this year, the annual tech summit in Paris and I was suprised to see many startups also focusing on physical AI. Basically take the LLMs and build a hardware product around it. This is a very simplified take but there were more advanced products, some also involving quantum computing. This is something I’m reading from tech CEOs too about how hardware is the next big thing. The opportunities seem to be massive and no wonder OpenAI is collaborating with Jony Ive (Ex-Apple) to create a device. We’ve seen quite a few products being launched already like the Rabbit R1, Friend necklace, Humane AI pin but none of them really clicked. We even have an AI-powered toothbrush now. Its not the same as software. Building hardware products is very hard but if you can crack the problem statement then the potential is massive.

I was talking to a professional in the semiconductors space and he says the current hardware systems are not equipped to handle LLMs. The battery for example is not optimzied for running these models. However, you don’t see the big companies trying to build small but capable models as the incentives are not strong enough. Its always about the best,biggest and smartest model that can rank #1 in LLMArena. Hugging Face in this case, is a nice example of trying to promote these smaller, application based models.

Think about it, an LLM on your smartphone or your watch or your electric car or just a camera attached to an LLM. Language translaction for example can be done on your device. There is a use case where on-device is better than cloud. More privacy, real-time processing but the intelligence is a tradeoff but I’m sure it will get better over time.

Bottom line, robotics, electric vehicles, smarter harware products have huge potential and pivoting to this can be pay dividends in the future in my opinion.

Photo by Alexandre Debiève on Unsplash