The week borrowing got costly and a model just decided

1. Borrowing got expensive again

The question

• Money got costly this week, the first Fed hike since 2023.
• The AI build leaned on years of near-free borrowing.
• Raising the cost cools prices but drags on growth. What gives way first, and how much lands on India?

My take

• Confirmed: the cheap-money era is ending. The whole AI build was funded by it, so the boom could hit real trouble from here.
• End of the bubble? Too early to say. But I am watching it hard.

Why I think so

• On 16 September the Fed raised rates a quarter point, to a 3.75 to 4 percent range. Its first rise since 2023.¹
• The vote was unanimous. Chairman Kevin Warsh pointed to inflation that will not come down.²
• Higher rates are how a central bank cools rising prices. They also make everything built on borrowing cost more.
• The Fed’s own dot plot shows 16 of 18 policymakers expect at least one more hike this year.⁹
• The Dow fell 631 points on the decision.⁴
• The 10-year Treasury bond moved back near 5 percent, its highest in years.²
• The safest asset on earth now pays about 5 percent with no risk. Foreign money sells Indian shares to go get it.
• Foreign investors have pulled about 2.41 lakh crore rupees out of Indian shares in 2026, about 17,222 crore in September alone.¹⁰

What I’m watching

• Whether the AI trade cracks as money gets costlier, and whether the Fed uses that expected next hike.

2. The model that only decides

The question

• A new model called Jev has one job: make the decision.
• You hand Jev the problem plus the choices, and it returns the pick with a confidence score. TypeSafe launched it on 15 September.⁶
• Until now the big, costly models made the calls. What changes when deciding gets cheap, fast, and repeatable?

My take

• This is the shift. Once picking the right option costs almost nothing, every other AI, and every app, can lean on Jev to make its small calls at scale.
• A super-fast, super-cheap checker sitting under the big models is a different kind of building block.

Why I think so

• Jev is not a chatbot. You feed it a question and a set of options, and it returns one choice with a confidence number.⁷
• It handles yes or no, one option from a list, or a score on a scale.⁷
• Most answers land near 100 milliseconds, and nearly all under half a second.⁷
• Input costs about four cents for a million tokens, the chunks of text a model reads. The output is free.⁷
• The proof it is catching on: by its first full day, Vercel said Jev was the fastest-adopted model in its AI Gateway history, used by about 13 percent of paid teams, more than double any launch before it, GPT-5.6 included.¹¹
• Cloudflare, LangChain and Langfuse built it in within three days.¹²
• Vendor tests claim workflows up to 190 times faster and 440 times cheaper. Those are their own numbers.¹²
• The founder, Diogo Almeida, is a former OpenAI researcher who worked on ChatGPT.⁶
• They call it a “System One” model, the fast gut part of the brain. The slow thinking stays with the big models.⁷

What I’m watching

• Whether more tools start routing their decisions through cheap classifiers like this one.

3. Worse and pricier at the same time

The question

• A product you relied on slides downhill as the bill goes up.
• The company stopped growing, so it switched to squeezing what it has.
• Real signal of a peak, or just prices catching up?

My take

• The growth is gone, so now they extract. That is enshittification in one line.
• The full breakdown gets its own reel this week.

Why I think so

• Swiggy’s platform fee was zero in 2022, then 2 rupees when it began, and is 17.58 today, its fourth hike in seven months. Zomato is about 15.¹³
• The Competition Commission ruled these fees are not an abuse of dominance, so the squeeze has a clear runway.¹⁴

Next week

• I pick these back up and see what moved.