There's a number i can't stop thinking about.
nearly 88% of all ai startup funding in 2026 has gone to companies based in the united states.
not 60%. not 70%. eighty-eight.
which means everyone else on earth — every founder in bengaluru, berlin, são paulo, lagos, tel aviv, singapore — is splitting the remaining twelve percent. i am one of those people. i'm 19, i dropped out, and i'm building a b2b company from indiranagar while a handful of labs in san francisco raise more money in a single afternoon than my entire addressable market is worth.
i want to be honest about what that feels like, and then i want to be honest about why i don't think it's the disaster it looks like.
The numbers are genuinely absurd
AI startups took roughly $202 billion in venture capital in 2025. that's close to half of every venture dollar deployed on the planet, and up 75% from the year before.
openai and anthropic alone absorbed something like 14% of all global vc. two companies. fourteen percent of everything.
anthropic is now reportedly the most valuable private company in the world — around $965 billion post-money after a $65 billion round in may. openai sits behind it at ~$852 billion and is reportedly lining up an ipo at close to a trillion.
q1 2026 saw roughly $300 billion in new startup investment globally. more than double the quarter before it.
read those numbers again and then remember: 88% of the ai slice landed in one country.
this is not a funding boom. it's a funding funnel, and it has a very specific shape. capital is not spreading out. it is concentrating — into fewer companies, larger rounds, and one geography.
What that actually means if you're not in that room
the tempting reaction is despair. the other tempting reaction is cope — some line about how india will "build its own openai." both are useless.
here's the read i've actually landed on:
The foundation model layer is closed. it's done. it closed while most of us were still reading about it. you cannot enter a category where the table stakes are ten billion dollars of compute and the incumbents have sovereign wealth funds on their cap tables. anyone in india telling you they're raising a seed round to build a frontier model is either lying to you or to themselves.
but the layer above it has never been more open. every dollar poured into those labs makes intelligence cheaper for me. the models i build on top of got dramatically better and dramatically cheaper while i wasn't paying for a single gpu. that capex was, functionally, a gift. someone else spent a hundred billion dollars building the road i drive on.
the question was never "can i compete with openai." the question is "what did openai just make possible that wasn't possible eighteen months ago, and who has a budget line for it."
that's a question a 19-year-old in bengaluru can answer as well as anyone in menlo park. maybe better, because i'm not distracted by the possibility of raising at forty times revenue.
the part nobody says out loud
there is a version of this concentration that is actively good for founders like me.
when money is abundant, everyone is your competitor. every mediocre idea gets funded, every category gets crowded, and you spend your first two years fighting eleven venture-subsidised clones who are giving the product away for free.
when money concentrates, the clones die. the ones burning $400k a month to look like a company run out of runway. what's left is a market where you have to actually be good, and where being good is enough.
i'd rather build in that market. i think most honest founders would.
and the valuations, since we're being honest
the median ai seed pre-money is now around $17.9 million. seed-stage ai startups get valuations roughly 42% above their non-ai peers. most of the category trades somewhere between 10x and 50x revenue.
i've spent three years around markets. i've watched what happens to a multiple when the story stops working. and i know exactly what a 40x revenue multiple on $2m of arr is: it's not a valuation, it's a promise. a promise that you will grow into a number someone else picked for you, on a clock someone else started.
so when i price my own round, i'm not trying to win the headline. a high seed valuation is a loan against a future you haven't earned yet, and the interest is paid in your series a — either as a down round or as eighteen months of pretending. i'd rather raise at a number i can grow into, keep the cap table clean, and never have to explain to my team why the story changed.
that is an unfashionable thing to say in 2026. i think it will be a very fashionable thing to have said in 2028.
both altman and goldman have publicly warned about bubble dynamics. mit sloan opened its year by asking not whether the bubble deflates but how fast. when the people holding the balloon start describing the balloon, you should probably listen.
the labs with real revenue will survive a correction. so will small companies with real customers and low burn. the casualties will be the middle — the wrappers priced like frontier labs, funded by people who confused a sector with a business.
So, twelve percent
i don't get to change the 88%. neither do you.
what i get to change is whether i'm building the kind of company that needs a bubble to make sense.
i don't have a $965 billion valuation. i have customers who have a problem, a co-founder who can build, and a runway i control. that's not a consolation prize. in about eighteen months, i suspect it's going to look like the only sane way anyone was building at all.
the money went to america.
fine. the problems didn't.