Portkey.ai sold to Palo Alto for $140 million ten weeks after raising its Series A. Its exit is worth celebrating but it also surfaces the expensive gap between building a global AI startup from India and keeping one independent.
In the fiction the company took its name from, a portkey is a seemingly ordinary object that magically transports whoever touches it somewhere far away, instantly.
In the same fictional universe, a horcrux asks for considerably more. You split your soul into pieces and hide them in vaults, hoping the fragments buy you immortality.
Rohit Agarwal and Ayush Garg named their startup after the first object.
In April, they used it as one.
American cybersecurity leader Palo Alto Networks agreed to acquire Portkey for $140 in total consideration.
This was roughly ten weeks after Elevation Capital led its $15 million Series A.
The deal closed in May.
It was an excellent and well-deserved outcome. Probably one of the first AI acquisitions from India that surpassed the $100m milestone.
Portkey had built one of the most credible AI infrastructure companies to emerge from India. It handled trillions of tokens every month and had found its way into the production systems of large global enterprises.
Palo Alto called Portkey’s technology a “mission-critical control plane” for autonomous agents.
Then, a few months later, another AI infrastructure deal supplied a comparison that seems awkward, for lack of a better word.
Stripe is reportedly acquiring OpenRouter.
The price tag? More than $7 billion!
Admittedly, Portkey and OpenRouter operate in the same domain but differ substantially in terms of scale and scope.
But the comparison is useful for another reason.
Portkey had raised roughly $18 million before selling for $140 million.
OpenRouter had raised roughly $175 million before its reported deal.
One company had much more capital available to find out how large it could become.
That distinction is something that we need to keep in mind when evaluating the state of Indian AI.
India has learned how to finance the creation of portkeys. Its next challenge is financing the horcrux.
The Series A That Never Got To Work
Series A capital is supposed to buy time.
Portkey’s Series A? Roughly ten weeks.
Elevation led the $15 million round in February 2026. Palo Alto signed the acquisition agreement on April 30.
The money barely had time to acquire a taste for enterprise software.
By then, Portkey had already done most of what we ask an Indian AI startup to prove.
Portkey’s technology was real. The customers were global. The usage was substantial. It had built serious infrastructure from India and sold it to global enterprises. Its product occupied an increasingly important part of the AI stack.
The Series A was supposed to fund the next question. An opportunity to test the next stage.
Portkey could expand distribution, deepen the product and find out whether an emerging infrastructure company could become a large independent platform.
How large could this have become?
We never got to find out.
Just to be clear, the argument is not “Portkey should not have sold.”
In three more years, Portkey might have become a billion-dollar company.
It might also have discovered that $140 million was the best offer it would receive.
A crowded market, compressed margins, an open-source substitute and an acquisition offer with fewer zeroes.
The founders had $140 million of hard cash on one side of the table and a probability distribution on the other.
They chose certainty.
Most people would.
The people perhaps worth questioning are those of us applauding from outside the room.
What exactly were we offering them if they chose the probability distribution?
Portkey Had A Good Problem
Palo Alto’s interest makes sense when you look at where Portkey sat.
As companies deploy more models and agents, they need infrastructure between their applications and the intelligence underneath.
Portkey provided that layer.
Companies could use it to observe model usage, enforce policies, manage costs and govern AI traffic.
Every month that AI usage grew, that position became more important.
It also became more interesting to companies like Palo Alto.
A security company looking at a future full of autonomous agents eventually has to decide how much of the control layer it wants to own.
Palo Alto wanted a control layer for securing enterprise AI. Buying Portkey gave it one.
There is plenty of precedent in infrastructure startups.
Google acquired Apigee. Salesforce acquired MuleSoft. Red Hat acquired 3scale. Intel bought Mashery.
Infrastructure has this peculiar habit of becoming strategically indispensable shortly before somebody larger decides it would look better on their balance sheet.
Portkey reached that moment unusually early.
This is a good cue for founders to help choose where to build.
A product can become more strategically valuable while simultaneously becoming more attractive for a platform to absorb.
A chokepoint can become enormously valuable.
Unfortunately, everyone around the chokepoint can see that too.
$140 Million Looks Different From The Other Chair
There is no doubt that the Portkey acquisition deserves to be celebrated.
Two Indian founders built infrastructure that became important enough for a global cybersecurity company to acquire.
Employees received liquidity. Investors got an exit. The founders created a meaningful company in a difficult category.
For Agarwal, Garg, their employees and their investors, $140 million is a serious outcome.
For Palo Alto, the arithmetic looks a bit different. A company valued at over $300 billion paid $140 million for technology it described as important to securing the agentic enterprise.
Portkey faced a much larger relative decision.
It could accept a certain outcome or spend several more years competing in a volatile market. Refusal meant hiring more people, fighting more competitors, raising more money and spending several years finding out whether the market was as large as everyone hoped.
That alternative was potentially valuable.
But it was also expensive. And uncertain.
Which brings us to the horcrux.
The Horcrux Is Capital
Venture capital offers founders a strange bargain.
You give away part of the company today so the rest of it has a chance to become worth much more tomorrow.
Then you do it again.
And again.
Every round puts another fragment in another vault.
In exchange, the company gets more life.
A founder rejecting a $140 million acquisition after Series A may need $50 million eighteen months later.
Paradoxically, success can make the problem larger. The next round might need to be $100 million.
The founder therefore needs something more tangible than confidence.
They need to know those cheques can exist.
We tend to describe founders who reject acquisition offers as being unusually ambitious.
But perhaps, that gives personality too much credit.
Somebody has to fund Monday.
Then payroll.
Then the American sales team.
Then another year.
Then the next round.
“No” is a very expensive word.
The OpenRouter Number India Should Measure Itself Against
This is what brings to the OpenRouter number that is useful.
A few months before the reported Stripe transaction, OpenRouter raised $113 million at roughly a $1.3 billion valuation.
That brought its total funding to around $175 million.
Portkey had raised about $18 million.
Let’s forget $7 billion for a moment.
The number Indian investors should stare at is $175 million.
That money bought OpenRouter something valuable. Something that might not seem obvious at first glance.
Not time. Optionality.
The optionality to survive competition and market downturns. The optionality to discover what the company looked like at a different scale.
It also changed what an acquisition offer had to compete against.
You could even argue that the reason why Stripe paid such a large premium for OpenRouter compared to its last known valuation was that it was buying up that optionality and pricing it as a function of its own valuation – for a company worth above $150 billion, paying roughly 5% of this valuation to secure an important piece of an emerging strategically-important market was a perfectly acceptable trade-off. Contrast that to the difference between the relative valuations of Portkey and Palo Alto Networks.
Now, there is no sensible argument that another $145 million would have turned Portkey into OpenRouter.
For all you know, it might have bought Portkey three expensive years and a worse ending.
But $175 million behind a company creates a different kind of optionality from $18 million.
It buys quite a lot of “not yet.”
Don’t Fall In Love With The Model
There is a second lesson hiding inside OpenRouter, particularly for founders currently adding “AI router” to pitch decks.
By itself, routing may be less interesting than the hype suggests.
Sometimes routing means selecting between providers serving the same model.
Sometimes it means choosing an entirely different model.
For sophisticated agents, some of that intelligence may eventually live inside the harness, where the system already knows what the task is, what has failed and what the tools returned.
OpenRouter becomes more interesting when you stop thinking about the router.
Think about MakeMyTip instead.
You can book directly from a hotel.
That did not make MakeMyTrip irrelevant.
Fragmented supply created an opportunity for someone to aggregate the market.
OpenRouter is making a version of that bet on intelligence.
If AI settles around a handful of dominant models, that position becomes less exciting.
If the future contains thousands of specialized models consumed by billions of agents, someone will aggregate that supply.
OpenRouter already has developers arriving on one side and models and inference providers arriving on the other.
That could become a market.
Portkey offered something similar but for a different audience.
Its natural habitat was the enterprise. Governance, observability and control became more valuable as companies deployed more AI.
That made Portkey useful to its customers.
It also made it useful to companies like Palo Alto.
A company that owns an expanding market can become progressively harder to buy.
A company that fills an important hole in a larger platform may become progressively harder not to buy. Especially if the cost of acquisition is a small price relative to your own valuation. A little over a rounding error.
India Is Getting Good At The First $10 Million
There is still much to like about the Indian AI funding environment today.
Good founders can raise seed rounds. Even large ones in the range of $5-10 million.
Global customers get attention.
A technically serious company with usage can raise meaningful capital.
Portkey is evidence that this machinery works.
The worry is that sometimes the machinery works too well.
A startup raises $15 million.
The product takes off.
Global customers arrive.
Then a $100 billion company puts a nine-figure offer on the table.
The investor who told the founders in March that they were building a generational company now has to tell them in April to reject $140 million.
And then write, or find, the next $50-100 million.
The more a company succeeds, the more expensive does conviction become.
On the other hand, India’s capital stack becomes thinner at exactly that point.
That is the gap that we need to be bothered by.
The Companies We Keep Long Enough To Meet
There is a perfectly respectable future in which India produces dozens of AI companies that sell to global technology platforms.
Founders get wealthy. Employees get liquidity. Investors return funds. Engineers go on to start more companies.
There is nothing about that future that is a failure.
But there is another prize available.
India could also produce companies whose value keeps compounding here long after the first attractive acquisition offer arrives.
We already know the country can produce the engineering.
Portkey proved it can produce infrastructure global platforms want to own. They also proved that you can run a global GTM from India – much like the canonical SaaS firms of the previous generation, Portkey’s customer base was largely in the US and for all intents and purposes, being domiciled in India wasn’t a disadvantage from a sales perspective.
The missing ingredient is patience… attached with a bank account.
Postman provides one glimpse of what that can look like.
Its enormous developer distribution belongs to the company itself. That has allowed value to accumulate inside Postman rather than making it merely a useful component of somebody else’s stack.
Zoho arrived at independence through another route entirely. It removed much of the external financing clock from the equation.
Obviously, neither path can simply be copied.
But they remind us that companies become independent through structure, capital and distribution long before independence becomes a declaration in a board meeting.
So What Are We Willing To Pay For?
Somewhere in Bengaluru right now is a company whose first serious acquisition offer will look enormous to the founders even if they seem almost trivial to the buyer.
The LinkedIn posts will arrive quickly.
There will be photographs, congratulations and a neat calculation celebrating investor returns.
All of that is well-deserved.
But the more important moment would have happened earlier, behind a closed door.
The founders will have looked at the number.
Then they will have looked across the table.
Can they keep building?
Is the next $50 million there?
Will another hundred be there if the first fifty works?
Are the investors who believed when belief cost $15 million still believers when it costs ten times as much?
Portkey had a buyer willing to put a price on leaving for a magical place.
India’s next great AI company may need investors willing to put a price on staying in India.
The portkey already has a price.
India now needs to put a credible price on the time required to find out what happens when a founder chooses not to touch it.
We know what the portkey costs now.
We still do not know what we are willing to pay for the horcrux.
