2026: Tech will be slightly down as the AI bubble will soft land.
Gold: Steady but slightly up in Q1 & Q2, dips somewhat in Q3, steady in Q4.
Silver: Gains
Oil: Same as Gold.
BTC: $100K–$150K. Crypto will pick up.
Not financial advice.
If you look at humans doing recursive activities, there is a spark of imagination and inspiration that improvises each iteration. Without those external inputs, recursive tasks will stop. Like the Powerpuff Girls, there is sugar, spice, and everything nice; that everything nice is outside our subconscious, which machines cannot do.
This is how the future of AI will be:
> Not seeing any breakthroughs in the next 3-5 years, as there is not enough $$ for R&D; all money is like that multiple Spidermen pointing at each other.
> Data centre and cooling water: this is a huge issue
> AI will be pushed to local
Sarvam will be India's only Sovereign AI, but they must be really careful, as they will face challenges with funding and GPU access, as well as other complexities of running a business in a highly controlled location. There are other important challenges, but I am not going to go into them; there are many experts on that.
Codex browser use is the best but still needs some improvement; it struggles with React button clicks, often tries to call the URL directly using a tool it wrote and fails with HTTPS connections. It also fails with burger buttons in the UI.
Three Predictions:
1. Some form of AI, probably neurosymbolic in nature, will come that is far more economical and data- and energy-efficient than LLMs, and it will make an absolute fortune.
2. LLMs, on the other hand, will never be all that profitable (aside from the chip companies selling shovels in the gold rush).
3. Today’s gigantic bets are premature, and most won’t pay off.
> Be AI companies
> provide cheaper accounts for devs to use
> train on their work
> offer it as an enterprise service at full price for AI.
This flow won't work with high cost.
My take is that large enterprises should host their own cheap OSS model.
There is no team Claude or Codex for me; everything depends on what my customer wants. For some, I use Claude; for others, I use Codex. I have re-subscribed to Cursor and am pretty impressed by their composer model. There are some use cases for which I use a local Qwen. As a business, whatever works.