UDIT AGARWAL (@uditmo2006) header
UDIT AGARWAL (@uditmo2006) profile picture

bowing to the hand of destiny

Joined July 2009
Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)Photo by UDIT AGARWAL (@uditmo2006)
📖📖
Our VP of content has more engineering in him than most engineers have content in them ;-) Fantastic breakdown on how a device can still hear without a mic.
13
Woah!!!
The aviation industry is the archetypal target for cyber attacks. The entire industry runs on one old, brittle system (FAA air traffic control) that is maximally safety conscious. Degrade one core component and it doesn’t just slow the system down, it forces a halt because they cant take the risk. This industry is a sitting duck in the new AI cyber risk paradigm. It will continue to be a target.
30
I went on State Street's markets podcast to explain why cybersecurity is now a systemic risk to markets & our entire digital system. I also get into regulatory risks, & why the prisoner's dilemma at the center of AI development means a pause isn’t coming. podcasts.apple.com/us/podcas…
What a pleasure it was to meet and speak to @michaeljmcnair on #STTStreetSignals this week, covering his recent work on #cyber threats, the #AI Prisoner’s Dilemma and potential systemic risk for markets. Not to be missed podcasts.apple.com/gb/podcas…
1
5
23
3,947
𝗜𝗻𝗱𝗶𝗮 𝗿𝗮𝗻𝗸𝘀 𝟯𝗿𝗱 𝗴𝗹𝗼𝗯𝗮𝗹𝗹𝘆 𝗶𝗻 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗽𝘂𝗯𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 𝗶𝗻 𝗯𝗼𝘁𝗵 𝗾𝘂𝗮𝗻𝘁𝘂𝗺 𝗰𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 𝗮𝗻𝗱 𝗾𝘂𝗮𝗻𝘁𝘂𝗺 𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀, according to the @MIT Quantum Index Report 2026. Only China and the United States are ahead of us in publication output over 2021 to 2025. The citation data give a more nuanced picture. India accounts for 6.1% of quantum computing publications, but 3.5% of the world’s top 10% most cited papers, placing us fifth on this measure. In quantum #communications, the gap is wider. Our share of publications is 5.6%, while our share of the most cited papers is 2.3%, placing us ninth. The research h-index rankings show a similar pattern, with India joint 9th in quantum computing and 10th in quantum communications. Patent activity has grown rapidly. #Quantum technology patent filings in India increased from 226 in 2023 to 883 in 2025, nearly quadrupling in two years. This was the fastest growth among the locations tracked in the report. India’s participation in quantum research is substantial, while its citation impact is stronger in computing than in communications. I believe, greater research influence would make our position more significant. I am very happy to see India occupy a place among the 𝘁𝗼𝗽 𝘁𝗵𝗿𝗲𝗲 𝗽𝘂𝗯𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗰𝗼𝗻𝘁𝗿𝗶𝗯𝘂𝘁𝗼𝗿𝘀 and the 𝘁𝗼𝗽 𝗳𝗶𝘃𝗲 𝗰𝗼𝗻𝘁𝗿𝗶𝗯𝘂𝘁𝗼𝗿𝘀 𝘁𝗼 𝗵𝗶𝗴𝗵𝗹𝘆 𝗰𝗶𝘁𝗲𝗱 𝗾𝘂𝗮𝗻𝘁𝘂𝗺 𝗰𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 is encouraging. We now need 𝗜𝗻𝗱𝗶𝗮𝗻 𝗶𝗻𝗱𝘂𝘀𝘁𝗿𝘆 to participate more actively in translating these research gains into useful quantum technologies and products. I congratulate the 𝗡𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 𝗠𝗶𝘀𝘀𝗶𝗼𝗻 for this achievement and I am happy that NQM now has a substantial research base to build upon. You can download the full report from the MIT page or from my LinkedIn post: lnkd.in/p/gznUaVf6 #NationalQuantumMission @PrinSciAdvGoI @IndiaDST @DrJitendraSingh @nqmdst
11
78
5
325
9,493
In my job I'm constantly learning from experts and deciding whose judgment to trust. The best filter I've found is to distrust anyone who is completely certain about something too complex for anyone to be certain about. They’re the most dangerous person in the room.
19
16
1
179
12,786
📖📖
When it comes to scandals, I have learnt over the years that: - in France, the scandals are always about money - in HK, the scandals are always about illegal building structures (dug out basements, verandas on terraces…) - in the UK and the US, the scandals are always about s.x I guess that for the media, it’s easiest to rouse up people’s anger and passions about what they lack in their lives?
1
18
Some of India's finest are also some of India's youngest. This small group (of a much larger and very vibrant ecosystem) has raised 340m+ much of it in the last 18mos or so. The founders are more jagged than ever. The investors need to catch up!
18
36
2
430
30,682
UDIT AGARWAL retweeted
Anticipating this is exactly why I spent so much time writing this report. This isn't a partisan point. People across the political spectrum share the concern, and the instinct is understandable. If AI poses a serious risk, pausing it feels like the responsible thing to do. But a US pause doesn't pause AI development. It pauses American AI development. The models still get built, just by China. And since you need better AI to defend against AI, it leaves us less safe. This is one of the biggest decisions in our nation's history. So I mapped out exactly why the race can't be paused, and where it's headed. It's probably the most important report I've written.
AI development will not be paused. It will continue to accelerate, not because the actors are reckless or naive, but because they are trapped in a prisoner’s dilemma that makes acceleration the only rational choice.
11
13
4
90
9,804
📖📖
AI regulation isn't going to stop at data centers. You might think Trump will just veto any regulation. But the labs don't just need to avoid regulation. They need federal legislation that stops a state court, especially in California, from holding them liable for damage done by someone else using their models. If, and when, a ruling like that goes the wrong way, it will make US labs uncompetitive and crush the US funding advantage. We will effectively hand China the AI race, and with it the most important advantage in history. Regulating our labs does not make us safer. It actually makes us less safe, because you need better AI to defend against AI. The models still get built, just by our adversary. Europe has regulated itself out of the AI race, and it has made Europe less safe. It is still connected to the global internet, so it is just as exposed as everyone else. Except it doesn't have access to the now gated frontier models it would need to harden its digital infrastructure. It's hard for people to accept that AI poses such a risk and we aren't regulating the labs. The problem is that you can't regulate the Chinese labs. So you just put your own labs at a massive competitive disadvantage, without even reducing the risk. It's far from an ideal situation, and it's hard to accept the reality we find ourselves in. But ignoring reality will lead to a worse outcome. Economist Harold Demsetz called this the nirvana approach. It means judging a real-world arrangement against an idealized alternative that couldn't exist under real constraints, rather than against the alternatives actually available. He argued instead for comparing real options with each other. And the reality is that if we overregulate the labs, or hold them liable for what others do with their models, US labs lose and we aren't any safer. We're actually less safe. This isn't a political take. There's enough of that on here. I'm just explaining how this plays out.
25
UDIT AGARWAL retweeted
India is building new factories, modernising old ones & entering industries it barely touched a decade ago. For young Indians ready to learn, build & solve real problems, manufacturing may be the biggest career opportunity hiding in plain sight.
16
11
3
68
6,292
📖📖📖📖
Institutional integrity is critical to allocation models—low integrity spikes unpredictability. Yet the Street cites AI & valuations. On our Watchdog Index, India is at record lows vs history & EM. See Link: tinyurl.com/3jcb8w98
11
📖📖
BREAKING: Alphabet, $GOOGL, Meta, $META, Microsoft, $MSFT, Amazon, $AMZN, and Oracle, $ORCL, have now disclosed a record $3.3 trillion in future spending commitments. These are off-balance-sheet commitments covering data-center spending, future equipment purchases, leases, and other obligations. $GOOGL accounts for the largest amount at $1.06 trillion, with its disclosed commitments surging nearly +700% over the last year. $META follows at $866 billion, with nearly half tied to data-center leases that have not yet commenced and payment periods extending up to 30 years. Meanwhile, Microsoft, $MSFT, has $593 billion in commitments, followed by Amazon, $AMZN, at $453 billion, and Oracle, $ORCL, at $322 billion. The AI infrastructure buildout is creating an unprecedented pipeline of future financial obligations.
25
I wish people had idea about these folks! It’s shameful to say the least!
When the foreign acquirer starts understanding how things work post acquisition in India. A decade ago Rocket Internet learnt this the hard way - had all the evidence against top management cheating them of 100crs. Tried legal way, went nowhere. Eventually they exited India. ‘Founders’ now chilling with 💰 - some are celebrated angel investors.
72
📖📖📖📖
Andy Mukherjee took an iPhone apart and counted the parts India imports. Go one layer deeper. Every one of those parts was cut, pressed or placed by a machine. A country’s machine tools are the closest thing economics has to a civilisational indicator. They set the ceiling on everything downstream: defence, chips, electric vehicles, aircraft. You cannot build what you cannot make the tools to build. Every workshop of the world had its mother machine. Britain had Maudslay’s lathe. America had its armories and Ford’s tool rooms. Japan had FANUC and Mazak. Steam got the credit. The lathe did the work. The world’s ten largest machine-tool builders earn about $21 billion a year, roughly a twentieth of Apple’s revenue. A small industry with an enormous lever. Nearly everything Apple sells passes through their machines. Japan has four of the ten. Germany has the largest. China has one. India has none. Twenty years as Apple’s classroom taught China to run machines by the tens of thousands. It never taught China to build the best ones. The tuition loop has a ceiling, and the ceiling is the mother machine. India founded Hindustan Machine Tools in 1953. The country remembers it for wristwatches. The watch unit closed. The machine gap stayed open. India buys about $2.2 billion of CNC machines a year and imports $1.2 billion of them, almost all of the hard 4- and 5-axis end. A nation that imports its machine tools imports its ceiling. The deepest import is the controller, the brain of the machine. India has always bought it from abroad. Owning the hand without the brain is not owning the machine. Ford said that without Johansson’s gauge blocks, his tool rooms would be blind. A country that cannot measure a micron cannot sell one. India pays for phones. Then for parts. Nobody yet pays for the machines that make the parts. Assembly is a job. Components are an industry. Machine tools are sovereignty. Count the parts in 2030. Then count the machines.
1
29
📖📖📖📖
There’s only one other time I’ve seen the market this complacent in the face of an imminent, systemic risk. And I’m having déjà vu, because the risk around AI related cybersecurity has the same shape and setup as the pandemic risk in early 2020. Though it’s arguably worse. Bearish market calls are a dime a dozen. But my threshold for making a public bearish call is extremely high. I’ve only done it once before…Feb 2020, weeks before the market fell 35% in a month. By early Feb 2020 it was already clear to anyone paying attention that COVID was going to be a global pandemic. We had access to a lot of experts, and they were unanimous in their belief that a global pandemic was inevitable. And yet the public and the market treated it like another Ebola or SARS headline. When I warned people to prepare, I was told that we couldn’t have a global pandemic in this day and age. Today, people have the same difficulty imagining that the internet could become unusable. It’s so fundamental to how we live that its continued functioning is treated as a given. In 2020, the market didn’t discount this risk until the evidence hit them right in the face. So the adjustment was violent and nearly instant. In modern markets big tops are typically a process. They don’t tend to crash from all time highs. Economic data slows, breadth narrows, credit widens, etc. The market takes the signal piece by piece and distributes over months. But this one fell out of a clear sky, because an exogenous shock doesn't have to generate those kinds of signals in advance. Quantitative models are good at extrapolating signals that have paid off in the past. That has increasingly become the backbone of financial markets, as well as betting and prediction markets. You can model a football game, or the weather based on data. But if a novel risk hasn’t shown up in the signals a model tracks, it doesn’t just estimate the odds poorly, it may not represent the risk at all. That’s why the market could sit at an ATH high in Feb 2020, with a catastrophe weeks away. And its why the market currently sitting near an ATH doesn’t give me much reassurance. For a legible, datable risk it can discount ahead of time. It bids up vol into an election or a Fed meeting. But it struggles to price a novel exogenous tail because it has no reference class and no event date. It stays silent, right up until the repricing. Today the cyber experts agree on the direction the way the epidemiologists did in February 2020. AI cyber capabilities now pose a risk to our entire digital system. And every day we get more evidence that we are nearing a tipping point. Our economy and livelihoods run on that system. The tail risk is a collapse, and its fatter than the market is pricing. I believe serious disruption is now a foregone conclusion, and only the magnitude and timing are still in question. The market may wake up to this risk because of a single event that impacts the public in a visceral way, but that won’t be the underlying problem. That’ll just make it clear that the rules have changed. This is a paradigm shift in what attackers can do and how many systems they can target. And like in 2020, the market can recognize the most obvious exposures without grasping how far the damage could spread. In the weeks before the 2020 crash the market did lightly discount the most direct, first order pandemic trades. The cruise lines, the airlines, a handful of others. But the fallout was never going to be contained to those names. You can argue that the market is already discounting some risk, as the AI related trade has come under strain, credit markets are tightening, and there are other red flags under the hood. It’s hard to differentiate how much of that is due to cyber risks (and its second order implications), rather than fears over capex sustainability, large capital needs, valuations, etc. but the important point is that the market as a whole is still not pricing the scale of the risk. Since the market tends not to slowly discount this type of risk, it means you can’t wait for confirmation. The opportunity is to position before the market recognizes what’s already qualitatively visible to experts. Which is what worked in February 2020. But this isn’t an argument for fatalism, because the setup gives us an opportunity. Protection against an unpriced tail costs very little, because no one is bidding it. If the event never comes, you’re out a small premium. If it comes, you get paid from an undiscounted base, which is where the largest repricings in history come from. That asymmetry means you only have to notice that the price of the insurance is cheap relative to the payoff. I should say that the risk to this strategy is timing. A cheap convex bet is cheap per unit time but if that event takes longer then you can get eaten up by the carry. I think the timing of the market realization has now approached the point where my own qualitative analysis suggests that it is appropriate to take steps to protect your portfolio. But explaining the reasoning behind that assessment would take pages. This isn’t financial advice. You should decide for yourself. I’m just making my own thoughts and logic public for anyone who cares how I am thinking at the moment. I've never hoped more that I'm wrong.
28
📖📖📖
It is very funny to see the latest psyop on Indian VCs, who are probably some of the smartest and hardest working people I know, and are largely in it for the game We'd probably make twice the much money with half the effort in public equities. And I'm fresh from a weekend, so let's dive in: a) There is very little genuine risk capital in India: Everyone loves talking, and preaching about the American VC model. But we want to see American outcomes without building American capital structures. Our funds are in years 4 and 6 of their 10+1+1 lives. Our portcos are developing alternatives to petrochemicals, finding oncology drugs, running the plants of major European pharma giants using their software, and building satellites to launch into orbit. Yet, almost every conversation with an existing or prospective LP eventually brings the same question in the first 15 mins: when are you returning capital? Nobody asks if our outperformers need additional growth capital, or there's an opportunity to participate directly in their next rounds. We want deeptech built over decades, and to finance it in a way capital will be returned before the next leap year. b) The natural pools of long duration institutional capital are largely inaccessible: American university endowments and foundations have helped build an entire ecosystem of patient capital. They can underwrite illiquidity over very long periods, supporting venture firms across multiple fund cycles. Indian charitable institutions face far more restrictive investment frameworks. This matters enormously. These are precisely the pools of capital whose liabilities and investment horizons are naturally suited to financing companies that might take 10-15 years to become important businesses. We want semicon cos, new drug platforms, aerospace, and advanced manufacturing champions. But where is the domestic institutional capital structured to wait for them? LPs want capital returned before the time for even one qualification cycle and a clinical trial is over. c) Domestic tech adoption exists mostly in press releases: Every Indian company talks about innovation, digital transformation, AI and working with startups. One of our portcos has won competitive global RFPs to deploy its software across vaccine and pharma sites of global top 10 giants. But struggles to find buyers for the same technology in India. Other companies in our portfolio can get demos and paid pilots quickly in SEA and EU. But in India only manages an MOU, a press release and then 12 months of silence afterwards. Most dont want to be customers of Indian tech preferring to pay outrageous sums for foreign players. There is a remarkable reluctance to become first serious customers of indigenous tech d) Secondary liquidity is a mirage: Conversations often begin with a substantial (often 50%) discount to the last funding round, followed by requests for additional governance rights and preferential terms. Understandable for companies underperforming. But those have no market. This 50% haircut only applies to the best performers, growing rapidly. For most emerging VC portfolios, a functioning secondary market is nonexistent. A surprising no of acquisitions are really acquihires dressed up as exits. The company gets acquired, the founders, and employees move on, and investors recover little or nothing e) Running a regulated venture fund is an increasingly expensive administrative exercise: Regulatory requirements keep evolving and compliance keeps accumulating. Reporting is literally an industry of itself. And every capital call becomes another round of explaining the same portfolio updates already covered extensively in quarterly reporting. Followed, inevitably, by the question about distributions. From a fund in year 4 of a 10+1+1-year life f) And actual capital deployment is extraordinarily time consuming: A $ 1mn cheque sometimes takes 6mos from sourcing to wiring. I've previously moved more money than that in public equities with two clicks, barely giving it a second thought. Of course, the work is part of the point. In venture, we underwrite incompleteness. But it does make the economics of the job amusing *g) All of this makes venture investing a consensus game: Every participant in the ecosystem is behaving rationally from their own position. LPs want distributions, enterprises want proven vendors, secondary buyers want discounts and downside protection, regulators want accountability, and GPs need to raise their next funds Put all these together and you get an ecosystem that increasingly rewards consensus investing. Companies with visible revenues (come back when you have X cr revenue), established customers (who is buying), familiar business models (consumer), recognisable co-investors (who else is ivnesting) and clear exit pathways find capital (where else will you get capital from). The ones attempting genuinely new things, with long development cycles, unknown commercialization timelines and no obvious comparables, struggle The irony is that venture capital is supposed to be the business of underwriting things before they become obvious. But we've invariably built an ecosystem where everyone wants someone else to take the first risk, validate the technology, establish the market and preferably demonstrate an exit before they participate. As Munger said "show me the incentive and I will show you the outcome", it is visible The strange thing is that none of this changes how fascinating the underlying work is. We get to meet founders building things that should not be possibe. We get to see technologies move from labs to factories and hospitals and orbit. Sometimes we get to be a small part of making that happen. I suspect that's why so many of us stay India has no shortage of people building with extra ordinary risks. She has an acute shortage of institutions who have the willingness to finance through the timelines required to deriskUntil that changes, we'll keep trying to build 20-year companies with capital that starts asking for its money back in year 5. But we'll keep doing it anyway Apparently, we really are in it for the game
25
📖📖📖📖
This is NOT a risk to gold prices and the reason why explains an important lesson about gold and bitcoin that very few people understand. There are obvious feasibility questions, but we can ignore those and even assume it pans out. Based on Marathon’s estimated goal, a large reactor could produce 5t/yr. So even if you build 100 reactors thats 500 t/yr, equivalent to 14% of global mine production. But mine supply isn't what matters. Other commodities are consumed. Oil gets burned, copper gets trapped in existing infrastructure, ect. So new production is required to meet continuing demand. But gold (and bitcoin) are unique bc they aren't consumed. Nearly all the gold ever mined still exists and is available to be traded. So the relevant supply curve isnt the flow of newly mined supply, its the willingness of the holders of the existing stock to sell at a given price. And even 100 reactors would only increase above ground supply by 0.2%/yr. Price doesn’t go up to incentivize new production. It rises (or falls) to a level where enough existing holders are willing to sell to meet the demand of buyers. Which is why the most explosive price appreciation in Bitcoin occurred when new coins were being mined significantly faster than today. Bc new issuance isn't driving price, its demand relative to the available stock. *The fact that gold isn’t tied to marginal cost dynamics is also why goldminers have often struggled to earn adequate returns. But that’s a story for another time.
34