Hello there, stranger! ๐
Here for a sip?
Your favourite tea from Sand Hill Road to Koramangala is poured, and this week, everyone's whispering about who is actually paying for AI.
Last week, I said the stock market was about to get a say in what frontier AI is worth. I didn't expect it to start this soon. This week, public investors began asking harder questions about the money behind AI: the debt, the spending and the growth needed to pay for both. Private investors, meanwhile, kept paying up. So is AI getting cheaper, or just getting priced properly?
Let's get to it.
For two years, investors funded AI's build-out with little argument. This week, they started worrying about what it costs and how much of it runs on borrowed money. The worry showed up in three places at once.
First, chip stocks. US chipmakers lost $1.64 trillion in value in just over a week. Reuters put it down to bets on an interest-rate hike and worries about the cost of building AI.
Second, CoreWeave. CoreWeave rents out computing power from AI chips it owns, and it bought those chips with borrowed money: about $50 billion of debt, against roughly $13 billion in sales expected this year. Its shares fell more than 15% in a single day, and its bonds began trading like those of a company in trouble. Investor Chamath Palihapitiya asked on X, "Is the AI buildout creating a private credit bubble?" Elon Musk replied with one word: "Yikes." The same day, Gartner warned that about four in five companies like CoreWeave could shut down by 2030.
Third, and most visibly, Anthropic. On Wednesday, Anthropic gave its biggest investors an update ahead of its IPO, Reuters reported. From July to September, it brought in $16.8 billion, 45% more than in the three months before. That's more than three times what it made in all of 2025. But in the previous quarter, sales had more than doubled. Growth is still fast, but it's slowing, and investors are also weighing the high cost of financing AI. Some now want to value Anthropic at about $1.5 trillion in its IPO, down from the roughly $2 trillion they expected. That's $500 billion less, before a single share has traded.
A week earlier, the Financial Times also reported that OpenAI brings in about $50 billion a year, not the $70 billion investors had assumed.
Investors still believe in AI demand. What they're now questioning is whether the debt and spending behind it will pay off.
There's a case for calm. Evan Schlossman of Neostellar Capital, a listed fund that owns CoreWeave shares, told Fortune that CoreWeave's customers are on "take-or-pay" contracts, which means they pay even if they don't use the capacity. And a lower Anthropic price may even help. Michael Parekh, who ran Goldman Sachs' internet research in the 1990s, wrote this week that "an IPO priced to sell, that trades well afterwards, is often better than one priced at the peak."
Even at $1.5 trillion, Anthropic would be valued at about 22 times what it now earns in a year, and well above the $965 billion investors paid in May. The question is no longer whether AI is growing. It's whether it can grow fast enough to cover what it costs.
The same week CoreWeave's bonds slumped, three private companies were valued far above their last prices.
Part of the gap is real. Modal and Ramp don't carry CoreWeave's hardware debt, so a pullback in AI spending would hit them differently. But part of it is timing. Public prices change every day. Private prices change only when a few investors agree on a new round or a tender.
Public investors reprice AI in days. Private investors reprice it only when they have to. If Anthropic lists well below $2 trillion, these private prices will be the next ones tested.
On Wednesday, three startups working on the physical side of AI raised money. Each round included a big industrial company.
Each of these giants is backing a technology its own business needs. Siemens, for example, sells electrical systems for data centres. Its executive Markus Grabmeier said, "Skeleton's portfolio perfectly complements Siemens' comprehensive data centre offering." Last edition, Samsung was doing the same with Indian chip startup Agrani Labs.
VCs invest for financial returns. Corporates also invest to secure technology and suppliers their own businesses need. With chip stocks falling, corporate money is starting to look like the steadier kind for hardware startups.
India could benefit. "Specialty chemicals demand has outgrown the models built to serve it," Radhika Ananth, managing director of India at Hitachi Ventures, told Global Corporate Venturing. She said much of India's factory capacity is underused and lacks the science and quality systems global buyers require. Scimplify is building a business to close that gap.
Zetwerk is a Bengaluru-based contract manufacturer, founded in 2018. Companies send it their designs, and Zetwerk gets the parts made through its network of factories, from pipes to aircraft engine parts. Its customers span oil and gas, renewable energy and electric vehicles. It's backed by VC firms including Lightspeed, Peak XV and Accel.
On 15 October, Zetwerk opened its โน5,012 crore IPO. At the top of its price range, the IPO values the company at about โน25,600 crore. In March, private investors valued it at โน27,000 crore. So the IPO price is about 5% lower, according to Mint. Mint notes the gap almost disappears if you leave out shares set aside for employees. Either way, there's no premium.
Even at that price, Zetwerk's investors want to sell more. The number of shares existing investors are selling in the IPO has grown by about half since the draft filing. Lightspeed, Peak XV, Accel and Kae Capital are among the sellers.
Big funds are buying. Before the IPO opened, they put in โน2,251 crore, including BlackRock, Norway's sovereign wealth fund and 22 Indian mutual fund companies. The wider market was cooler. On day one, investors bid for less than a quarter of the shares on offer. That's despite an improving business: sales in the June quarter were 55% higher than a year earlier, and losses shrank.
Tonbo Imaging, a Bengaluru defence electronics maker, opened its IPO the same day and drew far more demand. Its โน852 crore IPO is entirely existing investors selling their shares. On day one, investors asked for four times the shares on offer, even though Tonbo's revenue fell last year. Before the IPO, existing investors also sold โน413 crore of shares at the top IPO price to Indian buyers, including 360 ONE's defence fund.
Compare that with private funding. Indian startups raised just $113 million from private investors in the week to 10 October. That's 72% less than the week before. And next week, Jio Platforms opens a โน30,213 crore IPO. It's set to be India's largest ever. Right now, India's deepest pool of growth money isn't in VC funds. It's on the stock market, and much of it is Indian.
A slightly lower price you can sell at beats a higher one you can't.
In November 2025, the government launched a โน1 lakh crore fund to back research-heavy startups. It's called the Research, Development and Innovation (RDI) Fund. Last edition, chip startup Agrani Labs was applying for money from it. This week, the Economic Times reported who will hand most of that money out.
The fund picked its first 19 fund managers from more than 160 applicants. Together, they'll get โน7,131 crore. Fourteen are VC funds, including Chiratae, Blume, Kalaari and Speciale Invest. Five are research organisations, including centres at IISc, IIT Madras and IIT Bombay.
There are two conditions. First, the managers can only back companies at technology readiness level 4 or above. That's a 1-to-9 scale of how mature a technology is. Level 4 means its parts have been shown to work in a lab. Second, the managers must raise private money alongside the government's. The government's money also won't all arrive up front.
The government isn't picking startups. It's picking the people who pick startups. That's a sensible approach for a state that doesn't want to run a VC firm. And the matching rule means private investors have to say yes, too.
This is a start, not a flood. The first group gets about 7% of the full fund, and a second group is promised soon. There's a warning sign as well. The same ET report mentions an Indian Express investigation that raised conflict-of-interest questions about how another arm of the scheme picked projects. If the money moves on time and the managers find private partners, Indian deeptech gets patient money with a VC's discipline.
If you read one thing this weekend, make it this. It lays out both sides of the week's biggest worry. CoreWeave owes about $52 billion, but it also has $103.7 billion in signed contracts. The question is whether those contracts hold if AI demand cools.
Hello there, stranger! ๐
Here for a sip?
Your favourite tea from Sand Hill Road to Koramangala is poured, and this week, everyone's whispering about money that hasn't even landed yet.
This is one of the most fun weeks in venture in a long time. Not because everything is going up. Some things are going down, hard. It's fun because, for the first time in years, the market is about to answer the question everyone's been arguing about over dinner: what is frontier AI actually worth?
Let's get to it.
Anthropic and OpenAI have taken opposite paths. Anthropic is going public, so everyone gets to see its numbers. OpenAI is staying private and raising money from a small group of investors who agree on a price among themselves.
Anthropic's IPO filing, seen by Reuters, gives us the first real look at a frontier lab's finances. Revenue grew 12x in 2025, to nearly $4.6 billion. But spending grew faster. Anthropic lost $8.06 billion running the business. For every $1 that came in, about $2.75 went out.
It has also promised to spend $518 billion on cloud and computing power over the next ten years. Bloomberg says it could start pitching its IPO to investors as soon as the week of 9 November.
Critics flinched. AI skeptic Ed Zitron, CEO of EZ Primary Research, pointed out on X that Anthropic "spent $12.6bn to make $4.6bn in revenue in 2025", Yahoo Finance reported. I read the numbers differently, like those of any company building infrastructure before demand fully arrives. The spending always comes first. The real question is whether revenue catches up.
OpenAI took the other path. Sam Altman says there will be no IPO in 2026. Instead, OpenAI is reportedly raising at least $30 billion at a $1.4 trillion valuation from private investors.
You'd expect investors to prefer the lab that stays private and keeps marking itself up. But they don't. In September, tech newsletter Newcomer surveyed 25 VCs. Three in four said they'd buy Anthropic shares. Only two in five said the same of OpenAI.
So far, a small group of private investors has set the price of frontier AI. In November, the stock market gets a say. Newcomer called the coming weeks "very instructive in assessing the health of the AI business." Whatever public investors pay for Anthropic becomes the yardstick for OpenAI's $1.4 trillion, whether OpenAI likes it or not.
A price and a bid are different things. A bid is someone saying "I'd pay this." A price is someone actually paying it. This week brought a lot of very loud bids:
Zoom out, and the money is piling into a few giant deals. Crunchbase counted a record 27 rounds of $1 billion or more from July to September. Together, they took about a third of all the money startups raised worldwide. Indian money is part of it, too. Indian investor Premji Invest co-led a $145 million round in CScale, a US company that makes optical links for AI servers.
But a price only counts once someone can sell at it. PitchBook counted just 18 US IPOs in those same three months. Most of these valuations have only been agreed by the people already in the room. Anthropic is the first to walk out the door and find out what everyone else thinks.
Two software deals show how differently buyers now price companies.
Size can't explain the gap, because Airtable is the far bigger company. Salesforce paid for future growth. Bending Spoons paid for today's cash flow. A young AI product is priced on how fast it could grow. An older software company is priced like a mature business that throws off cash, which is exactly what Bending Spoons likes to buy.
For software founders, that's the game now. Revenue sets the floor. The growth story sets the price.
AI has three layers: the models, the apps built on them, and the customers who use them. The big question for the next five years is which layer keeps most of the profit.
Pranav Pai, founding partner at Bengaluru VC firm 3one4 Capital, thinks the AI labs are coming for the top layer: "The most profitable move in artificial intelligence may no longer be selling intelligence." His evidence: Anthropic launched its own design tool, Claude Design, after partnering with Figma. It launched Claude for Legal after featuring legal AI startup Harvey as a customer. And OpenAI now sells its own customer-support agent, after support software company Intercom built its agents on OpenAI's models.
Asset manager State Street Global Advisors sees it the other way. In a September note, it argues that the lead among AI labs changes hands so often that none of them can hold on to it. That leaves more of the profit for the companies using AI. Investor Tomasz Tunguz of Theory Ventures shows the pressure on the labs: the price of using AI models has fallen 41%, while usage rose 50%.
Both sides agree on one thing: owning a model alone is becoming a weak defence. If Pai is right, the safe ground is deep industry knowledge the labs can't see from the outside. If State Street is right, it's owning the customer relationship. Either way, the worst position is a thin product built on top of someone else's model.
India has more riding on this than most. Pai points out that Indian software firms are among the heaviest users of these models. So are the more than 1,900 global capability centres in India. In his view, that shows the labs exactly how their clients work.
India's biggest startup round from July to September didn't include a single venture fund.
Electric-scooter maker Simple Energy raised about $180 million. The round was led by the family office of A. Velumani, who founded the diagnostics chain Thyrocare. Another family office and a few angel investors joined.
Indian startups raised just under $3 billion in the quarter, less than in each of the two quarters before. This one round was about 6% of that.
Why would family money step in where VCs stepped back? Time. A VC fund has a deadline: it must return its investors' money within a fixed number of years. A family office has no such deadline. Rishabh Mariwala of Sharrp Ventures put it beautifully: "With a family office, it is a marriage without a divorce date." Building EVs takes a lot of money for a long time, so that patience is worth a lot.
And this isn't a one-off. The number of Indian family offices investing in startups grew from 45 in 2018 to about 300 in 2024, per Outlook Business. As VC money shrinks, Indian founders building capital-heavy companies now have a second, homegrown source of money. That's a big deal.
Agrani Labs builds computer chips and software to run AI in data centres. Four former Intel and AMD engineers founded it in Bengaluru in 2024. In February, it came out of stealth with an $8 million seed round from Peak XV.
Eight months later, it's reportedly raising โน800โ850 crore from Samsung, 360 One and Peak XV, the Economic Times reports. That's more than ten times its seed round. It's also applying for โน120โ130 crore from the government's Research, Development and Innovation (RDI) scheme.
Chips are hard to fund because companies need years of money before they ship anything. Agrani's answer is to share that load three ways: a VC, a large corporate and the government. It isn't alone. VC firms including Accel, Lightspeed and Bessemer have joined DeepVerse, a 30-member group set up to shape the rules for slow-to-build industries. It also feeds into the same โน1 lakh crore RDI scheme. If Agrani closes its round with the government money attached, other Indian deeptech founders get a template to copy.
Two IPOs, days apart, ended very differently.
Online lender Moneyview raised โน1,092 crore in its IPO. Investors applied for about 100 times the shares on offer. On its first day of trading, the stock opened more than 60% above its IPO price.
Snapdeal's parent company, AceVector, raised โน420 crore. Demand was weaker: investors applied for about five times the shares on offer. Then the stock listed 11.6% below its IPO price and ended its first day 18.4% down. Snapdeal was once valued at $6.5 billion, and it cut its losses sharply last year. But investors still passed.
Inc42's analysis explains why in one comparison. In FY26, about 12 million people bought something on Snapdeal. On Meesho, 264 million did.
The lesson for India's IPO pipeline: a famous name and a smaller loss won't carry you. Scale will. Three more companies are lining up to test that:
If you read one thing this weekend, make it this. It's the quarterly scorecard for US venture, and it tells this edition's story in two numbers. AI took a record share of the money invested, more than four in every five dollars. And only 18 companies went public.