Suyash Handke

There was a lot of talk in 2023 and 2024 about startups using a one-time payment model, but earlier this year, I started to see founders going back to subscriptions. What do you think?

Some products are just too early for the market, only to be rebranded or reinvented years later. Think of Google Glass, the Microsoft Courier, or even early versions of VR. What s a product you loved (or saw potential in) that was just ahead of its time?

The new AI model selloff cost the world s 500 wealthiest people $108 billion, whipping out $2 trillion worth of US stock market capital. On Monday, Nvidia alone lost $589 billion in market capital. Deepseek is China s new and very first AI model, and it made headlines within a week of its release. Now, we all see new AI models heading up, but Deepseek is what most of us didn t see coming. Instead, I am intrigued by its cost-efficient model, which is shaped using reduced-capability chips compared to its predecessors, Chat GPT and Gemini. Within a week of its launch, this Chinese app has become the most downloaded free app in the US. While OpenAI s GPT-4 incurred a training cost of $600M DeepSeek s V-3 model is claimed to have been created within a budget of $6M. Interestingly, what took tech giants like Open AI and Google years to build is now claimed to have been achieved in 2 months, in a thousandth fraction of its cost. If we compare the infrastructure, DeepSeek is comparatively stronger. With over 685 parameters and a cluster of Expert MoE architecture, it seems like a better choice at first glance. The investors in panic mode are getting communities buzzing about how investors never think long-term. As in this case, most of Nvidia s sales come from selling server systems, GPUs, and entire data centers with racks, networking, CPUs, etc. The panic mode among investors from a single release doesn t make much sense. Whatsoever, the AI revolution has come to another level and is evolving at a much faster pace! Here are some of my takeaways. Firstly, since Deepseek has attached a number against its development, do we need 500 Billion funds to achieve this? The answer is somewhere in the middle. While the infrastructure will continue to become cheaper and better, the race to the bottom is inevitable. However, this doesn t mean we can skip the cost of developing and figuring out the background architecture. A simplistic explanation is to treat AI as just another progressive step in Software. It's something David Friedberg has been advocating for a while. Second, even before we know what AI can do, the point of AI regulation seems like an overkill, and I think Deepseek's innovative development has somewhat made the point. (It doesn t matter which side of the fence you are on or whether it s a Sputnik moment triggered by sanction or just another AI meltdown) Lastly, we all should remember that it s a tool rolled out by a hedge fund, which by nature is known to bias the market in its favor. Only time will tell whether they want to run it as a business or it s just a way to short on US AI giants to win some extra IRR. I leave the politics for the journalists to figure out. Let me know what you all think. -Abhinav