2025-10-11 08:00:00
I have tweeted about this many times in the past. This is such a big opportunity that no one has nailed so far here.
Palantir won’t bother servicing companies in Southeast Asia. They remain focused on the West, mainly for cultural reasons and because of the larger deal sizes.
OpenAI, on the other hand, won’t send a full deployment engineering team unless the deal size is around $10 million (if I’m not mistaken). Most AI curious companies are just using these LLMs to chat, but have not got the most out of them.
People talk about building “Infosys 2.0,” but doing that requires more than just a team of smart techbros, you also need a founding team with the right connections to open doors. (Take a look at the composition of the Brainco team for example from the transcript below.)
I heard rumors that a VC firm was seriously considering bringing together a group of AI experts from Silicon Valley to launch something similar in India. But people are underestimating how difficult it is to sell large-scale transformation projects.
You need both capability and access.
Infosys and the other WITCH companies can’t pull this off. They simply don’t have the muscle for it. This will have to come from a completely different company altogether.
“TBPN: You’ve been working with Jared Kushner on a new company. I know he’s been busy, but I’d love to hear the latest.
Elad Gil: Sure. We recently launched a company called Brainco, which focuses on using AI as a platform to help transform the world’s largest institutions. We’ve been working with a number of large enterprises, private equity firms, and others. It’s been a fun project to build alongside Jared, Eric Wu, and Luis Videgaray. Eric was the former CEO of Opendoor, and Luis served as the Finance and Foreign Minister of Mexico. It’s an interesting mix of people coming together to solve some of the biggest AI challenges.
TBPN: So was this born out of realizing how much firms like Accenture or McKinsey were charging to make “AI strategy” pitch decks—and thinking you could do it better by actually building software for these organizations?
Elad Gil: It’s definitely more focused on the software side. We’ve built a common platform that helps enterprises manage different forms of data, evaluation frameworks, and the other core systems they need to truly adopt AI. On top of that, we build vertical-specific applications—and sometimes horizontal ones—that can be reused across similar companies in the same industry.
For example, in financial services, there are a dozen things every major firm needs to build. There’s always some customization, but the core infrastructure remains the same. Think about large enterprise deals—if you’re Dell, VMware, or Oracle, and you’re doing a $50 million contract, you’ll absolutely tailor parts of it to the client. We’re similar in that way: a shared platform and vertical solutions, with client-specific customization where it matters.
TBPN: How are you thinking about your target customers? Are you focusing on certain verticals, like avoiding healthcare because of HIPAA, or defense because of FedRAMP? Or are you segmenting more by company size—say, only working with large enterprises, not mid-market firms?
Elad Gil: The goal is to work primarily with the world’s largest institutions—those with significant revenue, market cap, or impact. There will always be exceptions, but we’re focused on helping organizations that can use AI to create real leverage at massive scale.
TBPN: That makes sense. There was a report that JPMorgan is spending about $2 billion a year on AI to save roughly the same amount through automation.
Elad Gil: Exactly. They’re basically breaking even—though some of those savings are likely durable—but it shows how much value and investment is already being driven by enterprise AI adoption. (Note: Transcript cleaned up using ChatGPT)”
2025-10-10 08:00:00
“There are about 150 people who run the world. Anybody who wants to go into politics, they’re all puppets. Okay? There are 150, and they’re all men, who run the world. Period. Full stop.
They control most of the important assets. They control the money flows. And these are not the tech entrepreneurs. Those guys are going to get rolled over in the next five to ten years by the people who are really underneath, pulling the strings.
When you get behind the curtain and see how that world works, you realize it’s unfairly set up for them and their progeny. I’m not saying that’s something we can rip apart, but the first order of business is: I want to break through and be at that table. That’s the first order of business.
The way I do that is by proving I can do what they do as well as they do it and then do it better. Because at the end of the day, they are commercial animals. And they’ll open the door out of curiosity, and they’ll let me stay because I add value. And once I’m there, I can open the door for other people who can try to do the same thing.
My entire goal now is to be in a position to aggregate enough of the world’s capital to then reallocate it according to my worldview. I’m not saying my worldview is the best or the right one, but it’s mine.
And at the end of the day, there are 150 other guys with their own worldviews, and they don’t care what you think about theirs. That’s the truth. So why not me? Why not one of you? Why not?”
Chamath Palihapitiya at Stanford Graduate School of Business
2025-09-17 08:00:00
The world has become extremely chaotic.
You do everything right, get that fat FAANG offer, and discover that your future now depends on how an immigration lawyer interprets a random H1B update. Your company announces a record quarter, but also 10% layoffs due to “AI efficiency”. There is no baseline pretection anymore.
The rules have become arbitrary. People are not built for this much uncertainty.
With the stable path collapsing, people are doing two contradictory things at once. On one side, they are retreating into things that feel lindy. Becoming conservative in values: more religious, more tribal, more skeptical of outsiders. Hatred for ‘the others’ is on the rise.
On the other side, they are getting more degenerate with their money. They are day trading options. Putting their savings into meme coins.
The middle class bargain was a duration trade. You exchanged years of predictable effort for a slow compounded growth in your networth. The safe path does not exist anymore. The middle class ladder is gone. The dream of security through hard work has collapsed.
And everyone is trying to escape the permanent underclass. With the normal way up blocked, people are just gambling harder than ever.
The archetype of the next decade is the conservative degenerate. This is the age of hyperspeculation.
2025-08-27 08:00:00
I have shared this in private with a few people. Unfiltered thoughts below.
India SaaS has always been hard. Compare the money that has gone into India SaaS with the actual outcomes for these companies. We still gave it our best. The advantage was price arbitrage. A lot of the cost was GTM and people. You could set up a cold outbound center here, hire a big engineering team here, start selling in India, then try to sell in the US. It worked for a while. But when the price of software shifts from twenty dollars per seat to replacing people, relationship driven sales start to matter a lot. In that case, you need to do more in person sales. And now the India salary arbitrage does not work either. Indian salaries have gone up. Instead of fixed API costs, you have variable foundation model costs, and a lot of your topline revenue going to these foundational model labs. So you cannot build another Freshdesk from India with the same playbook. Look at the revenue growth of Indian SaaS startups, how much funding they have raised in the last two years, and which startups are getting invested in.
Oh, but horizontal SaaS works, why don’t we try that? Actually, even in the US it is mostly AI SaaS, and the SaaS startups there are also getting gobbled by PE, listing at lower multiple, and most investments are now in AI. When every idea has 50 copycats, what matters is capital accumulation and narrative building. Say you do an AI design tool from India and raise 1 million from an Indian VC, you build in Bengaluru, and a startup with a similar idea and similar ability of the founding teams appears in the US, but your competitor is ex Figma, a second time YC founder, and manages to raise 10 million from a16z. Now a16z is adding them in their market maps, not you. Garry Tan is shilling them not your startup every week. Your Indian VC is now panicking. First they were saying grow fast, then only you will get your next round, now seeing you burn 3x of your revenue in Claude API costs, they are asking you to become profitable instead. The same schizophrenic behavior based on market cycles, except cycles are like monthly now, driven by narratives you cannot control. Do you think Cursor would ever get funded in India? No. Windsurf got exit liquidity thanks to Google. Being in the US helps. Network helps. So no horizontal AI SaaS? What do you do?
If the main cost for AI SaaS companies is foundational model cost, then the only ones that can survive without getting copied by the labs at their app layer or burning their runway are the companies that replace employees or cut opex. You cannot do seat based pricing where you charge $20 but spend $60 each month, hoping nonusers make your business model work like in a fitness gym. Power users in AI SaaS will always consume heavily and pick the best model, so churn will stay high and much of today’s revenue is still experimental, with a power law at work. If you switch to usage based pricing, you are essentially reselling Claude credits, and your burn ends up two to three times your revenue again, because power users will always choose the best model. So no seat based pricing and no usage pricing. You have to sell outcomes: do the same work with 10 people instead of 50, and pay us $2K per user per month. I am sure a $200 Claude subscription will become $2,000 soon. This is why vertical SaaS is where most Indian builders and VCs will focus: financial copilots, AI doctor, AI nurse, AI accountant, direct job replacement, so you can make $10K per month and break even. But you have to sell in the US. There is a reason Indian VCs tell founders to move to the US right after the pre-seed round. VCs in India are mostly investing in vertical SaaS for the US. In India, if you try to say sell an AI QA tool, companies will laugh you out of the room; they would rather hire fresher QAs than pay more for an AI tool. Indians just throw bodies at problems. We are not supply constrained. So you cannot have replacement-based pricing. The only option is moving to the US and selling there.
What works then? What should Indian founders build?
Not just AI. I think a lot of money and energy will flow into deep tech too. It is not a question of if but when.
In the last four months since I left my job, I have worked with or met at least 20 to 30 companies, and I can speak to the margin structure of most AI businesses. I have met enough VCs as well, and I think most would privately agree that this is an accurate read of the situation here in India.
2025-08-17 08:00:00
I’ve been a fan of Superpower since I read the founder’s blog on working through the healthcare idea maze.
Have followed the company closely ever since. Superpower has built a strong business in a crowded market. We can reuse parts of their approach (based on my study of the company) to build something similar.
The core loop is the Data, Trust, Advice flywheel.
Capture high-quality data, use a trusted agent to turn it into personalized insights, then recommend the next best action. Each accepted recommendation feeds back more data, sharpening both the model and the relationship.
The only things that matter initially are the quality of the data and the continuity of trust.
You need data that’s both valuable and hard to capture. Superpower chose health because health data unlocks outsized value. Choose one wedge and solve the capture problem better than anyone else. You can layer on everything else later.
Your agent must clear three trust gates. First, competence, mirror a user’s raw data back to them so your accuracy is obvious. Second, alignment, be on the user’s side, not an advertiser’s, so skip affiliate links at the start and charge a subscription. Third, reliability, show up every time, hit uptime targets, meet compliance, and run evaluations that catch mistakes.
Sequence monetisation with care. Most teams try to take value before they create it, bolting on marketplaces and revenue share before users trust their basic recommendations. This kills the flywheel before it starts spinning.
The right sequence starts with what I call the mirror moment. You need to prove that you can ingest and reflect someone’s world more accurately than any alternative. Spend as much time as needed getting it right. Track one metric: the share of data sources you connect and show on an interface without errors.
Then move to nudges. Surface low-friction, high-confidence actions, like swapping a snack or trying a peptide for a cold before it goes mainstream.
[The Superpower cofounder injected a peptide on a podcast host recently.]
The marketplace phase, then your own products, is where real money sits. It only works once the agent is genuinely loved. People will let a trusted advisor broker product picks and take a cut, but they’ll leave the second it feels like you’re pushing stuff they don’t need. You need to measure how often people actually follow your suggestions. If this rate is low, your trust hasn’t been earned yet. If it’s high, only then you can start thinking about becoming a trusted marketplace.
The marketplace phase is where real money lives, but it only works after users genuinely love the agent. People will let a trusted advisor broker product recommendations and take a cut. They’ll abandon anything that feels like it’s selling them stuff they don’t need.
Watch a few metrics. True Data Density, how much unique live info you collect per user per day. Advice Utilisation, who actually takes your suggestions. Outcome Delta, whether lives measurably improve. Referral Coefficient, how many users bring friends. And gross margins after advice delivery costs.
The moat isn’t just having the data. It’s built in layers. You need the frameworks that map raw inputs to actionable goals. That takes real expertise. You need behavioural history, knowing which nudge style each user actually responds to. That cannot be copied, even if a competitor captures the same data you have. Over time, you can even capture regulatory advantages. Get first-in-class certifications, then lobby to make them table stakes for everyone else. At every step, the moat compounds, but only if you’re patient and paranoid about breaking trust.
In a world where everyone’s drowning in choices and information, the most valuable thing you can offer is a trusted concierge that knows users better than they know themselves. With permission to act on that knowledge, you can build something massive.
2025-08-11 08:00:00
I stumbled on this podcast with the legendary investor Christopher Hohn a few weeks back. YouTube’s algorithm is magical sometimes. Hohn sticks to old-school investing. He skips tech and picks physical businesses with real assets. Some of his wins include toll booths and airports.
Though I lack the cash or connections to ever buy toll booths in Europe, I wanted to find modern versions.
If toll booths and airports are physical businesses where you get paid every time someone passes through, what’s some other toll booths for today’s age: something with recurring, unavoidable usage and strong barriers to entry?
A toll booth business sits at a critical choke point in a growing market, charging a small fee for every transaction that passes through. You spot these opportunities by targeting layers that meet four criteria.
They must be
How do you identify a modern toll booth business opportunity?
In software, toll booth economics show up in several forms:
These businesses work like airports or toll roads, generating predictable cash flows, have high margins once built, lock customers in for years, and can steadily raise prices over time.
Here are some new age tollbooth businesses: AI inference services that route traffic across models, skimming a fraction of a cent per token. Synthetic data exchanges certifying privacy-safe datasets, taking a cut of each license. Passkey identity rails that charge apps per authentication, leaning on security audits as a moat. Programmatic carbon clearinghouses matching offsets with buyers, clipping a few basis points per tonne. Autonomy OS licensing certifying drones, charging per mile monitored.
Approach to build a toll booth business: