2026 Letter: Intelligence, Scarcity, and the End of Easy
Among all of the different ways you can deploy capital, the future comes for venture first. Every new paradigm shift hits Sand Hill Road like a ton of bricks with some investors building generational careers and others having their generational careers wiped out. And every year between those cycles is a constant balancing act of measuring how far from the top, or bottom, we are. Are we catching a falling knife? Or sitting out the rebound?
For much of my admittedly short career as a venture investor, chasing certainty has been the way to generate returns. Over the course of two decades, you could create outsized returns backing the most repeatable, simple businesses in B2B software: low marginal costs, infinite distribution, and long-term rents charged to captive customers. Why invest in consumer when you have to worry about marketing risk and trying to catch a genie in a bottle? Why invest in hardware or biotech with technical risk and large CapEx requirements? Why invest in anything complex at all when you have fixed income-like revenue with equity-like returns in Software-as-a-Service?
That world no longer exists. Building software is cheap and plentiful with Lovable and Cursor. Automating workflows costs $100/mo with Claude Cowork. And presentations, spreadsheets, and even letters such as this are crafted with an AI co-pilot. AGI-lite is here.
In this letter, rather than looking backwards at fund performance or market shifts like previous letters I wrote at Horseshoe, I attempt to answer what I think will be the crucial question of the next decade: As intelligence becomes abundant, and resources become less scarce, who wins and who loses?
Intelligence too cheap to meter
“Science has a simple faith, which transcends utility. It is the faith that knowledge of nature will help us to control the forces of nature” - Vannevar Bush
The last couple weeks have been particularly entertaining to me. I, like Michael Burry with recessions, have predicted three of the last one (maybe?) end of software moments. And I’ve been saying this for 18 months. Then, all of a sudden, public markets had an epiphany. Every time that Anthropic or OpenAI announces a new tool, the collective market acts like the “legacy” company at hand has been handed a Wells notice.
First, Anthropic launched a legal plugin for Claude Cowork. Thomson Reuters lost almost $10B in market capitalization in a single day as a result. As if Thomson Reuters is just one very large, very fancy AI plug-in. Then, a <$10M karaoke company announced a new AI logistics whitepaper that caused CH Robinson, the largest freight brokerage in the world, to trade down 24% intraday. Imagine telling a bystander that markets are efficient! And in one last example, Anthropic announced Claude Code Security and businesses like CrowdStrike, Zscaler, and Cloudflare all traded down 5%+ in one day. It’s clear to me that public markets must not know what any of those companies do. Claude Code Security scans your code for vulnerabilities and helps fix them. If I owned a big chunk of a Shift Left company like Snyk, I’d be concerned. But, if I own a CDN platform or a zero-trust network security solution, Claude Code Security is just a non-event. It’s like betting against Nobu Matsuhisa because a robot can flip burgers at McDonald’s.
These comments are not to discredit the larger trend. The capabilities of today’s AI-assisted coding tools are unreal. Nine years ago, it took my team of four at Chromata almost a year to launch a context graph API. It took me a couple weeks to rebuild the same product with Claude Code. The big rate limiter left is cost. And that’s dropping at an unbelievable pace. When GPT-4 launched 3 years ago, it cost about $36 per million tokens. A year and a half later, GPT-4o cost about 150x less. Unlike energy in Lewis Strauss’ prediction, intelligence really is approaching too cheap to meter.
How? Step one is abundant compute. Naturally, everyone wants data centers. Hyperscaler CapEx is reaching $400 billion on AI infra annually. OpenAI, Oracle, SoftBank, and others are pouring $500 billion into building 10 GW of data center capacity by 2029. And just when you might think capital is running out, we find ever newer instruments to finance this mania. Meta is building Hyperion, a 2 GW AI data center with over 4 million square feet in Louisiana, in partnership with a consortium of investors led by Blue Owl and PIMCO. At ~$30B in total, it is the largest private credit financing ever.
It is still unclear whether this is justified. A critic would tell you that we’ve seen this movie before. The telecom industry spent $500B over five years to build out almost 100 million miles of fiber optic cable to satisfy networking demand. But, by the time the build out ended in 2001, ~95% was “dark fiber” or completely unused. 27 telecom companies went bankrupt by the end of that year and $2 trillion+ of telecom market cap was wiped out.
On the other hand, “dark compute” seems unfathomable to me. Just a couple years ago, a vast majority of AI compute was spent on training models. This year, circa two-thirds will be spent on inference. And, this is despite the average consumer still searching Google first and a fuzzy landscape of consumer AI applications. For now, it seems to me that the status quo is: if you build it, they might come? When the paradigm shifts in tech, that’s a bet you have to make all day.
The sovereign compute race
“The nation that makes a great distinction between its scholars and its warriors will have its thinking done by cowards and its fighting done by fools” - Thucydides
Amidst this digital infrastructure buildout, global chaos has spread. A potential regime change in Iran. Maduro was airlifted out of Venezuela in stealth. Denmark is dreaming of Marty Lipton for Greenland. Russia is still on the offensive in Ukraine. Israel is under threat from the north and south. Mexico is burning after El Mencho’s killing. And the ruling government of Japan, a country with “no army”, is floating nuclear armament. The first couple months of 2026 has reminded all of us that maps are not rigid; each line and label needs to be defended by a combination of soft power and hard weapons. And, as the world wakes up to this, many countries have decided, for the first time in nearly a century, that it is every man for himself.
This uncertainty has led compute, like energy in the 2000s, to become the key theater in global posturing. Nobody wants to be caught on intelligence the way Europe was on energy after Russia invaded Ukraine. France has committed €109 billion to their “AI Action Plan”. Saudi Arabia has launched a $100B AI initiative, Project Transcendence. China has competitive, if not market-leading, open source alternatives like Deepseek and Seedance for every major American model. Capital will only continue to accumulate to these projects globally as matters of emerging opportunity and national security.
And then there is India. Over the last 100 years, the world has been bipolar. The Allies vs. Axis in WWII, the Americans vs. the Soviets in the Cold War, and the USA vs. China for the last decade. Everyone else picked a side. But, over the last decade, India has charted a new path: a BRICS member that has refused to sanction Russia while also being the member of The Quad that was the first country to ban TikTok over Chinese tensions.
Fundamentally, India can’t be ignored. The country of 1.4 billion people is a top five global economy by GDP still growing 6.5% YoY. They are home to the largest pool of STEM students in the world and have 4.3 million software engineers (second only to the US). They’ve built in tax credits for foreign companies using Indian data centers and up to ₹40,000 crore (~$4.8 billion) in electronic components manufacturing projects in their last budget. Last week, New Delhi flexed this muscle at its AI Summit with the CEOs of Google, Anthropic, OpenAI, and more in one room. For the first time, India has a chance to live up to its promise and is a really exciting geo to invest in.
But, our global compute ambitions are still heavily constrained by the laws of physics. Namely, how we power all of these chips and infrastructure. The aforementioned 2GW Meta datacenter will consume enough energy to power ~1.5 million homes. Transmission capabilities are decades old. Getting a permit for a new power plant has already become a Herculean effort. Google pulled a bid to build a 468 acre data center in Indiana after hundreds protested a vote at the Indianapolis City-County Council. As America lets uneducated activists determine our compute and energy policy, the rest of the world stands to gain.
Yet, certain markets in the US remain keenly interested in solving these problems. Texas, for example, remains a key source of energy innovation. From the original pioneers in the Permian to modern wildcatters innovating on top of the deregulated ERCOT electricity market, the Lone Star state is a much needed haven for innovation. There are so many opportunities to build in the natural gas market, in power optimization, and in grid and battery modernization. I remain, perhaps irrationally, hopeful that nuclear will one day be accepted as a critical source of abundance as well. Energy doesn’t have its version of Moore’s law but it is, nonetheless, a foundational layer that must innovate to keep up with the paradigm of intelligence we are betting on as an industry. After all, we must remember, bits run on atoms.
Let them eat tokens
“Socialism never took root in America because the poor see themselves not as an exploited proletariat but as temporarily embarrassed millionaires" - John Steinbeck as often paraphrased
One of the peculiar ironies of political commentary is that socialists seem to do the best job of critiquing socialism. George Orwell has likely turned more people against big government than has Ayn Rand. And Steinbeck intended to discredit the anti-Communist agenda in America with the above quote. Instead, I think it points to a key part of what makes America so special: Americans believe the future will be better and are willing to bet on it, whether that be with labor or with capital.
However, over the last several years, that bet has unraveled. The perception, and reality, gap between blue- and white-collar Americans has meaningfully widened. US non-housing debt has crossed $5 trillion at the same time that Delta’s premium cabin revenue eclipses its main cabin for the first time. It’s no surprise that there is deep resentment and that there is an accompanying bipartisan rise of populism.
In fact, “rise” may be the wrong word. Populism is already here. Josh Hawley and Bernie Sanders, both Senators who seemingly agree on almost nothing, co-sponsored a bill to cap credit card interest rates at 10%. My current home city of New York elected a Democratic Socialist with record turnout on a platform that included free buses and city-run grocery stores. And Ro Khanna, a US House Rep that espoused the greatness of technology at the same conference as me just 8 years ago, has traded his Silicon Valley constituents to grandstand to Teamsters with the message that he is putting labor over capital.
AI is only accelerating these fears. For the first time, technology-driven job insecurity is not constrained to certain pockets of the economy. Khanna’s favorite phantom protected class, truck drivers, are going to have to contend with the mass automation of their jobs. At the same time, white-collar engineers in San Francisco and junior bankers in New York feel like large swaths of their jobs are being automated away. The NY State Senate now has a bill, S7263, that intends to ban “chatbots” from answering questions related to medicine, law, engineering, and much more. Yet, we as the broader tech industry keep promising an AI-driven utopia.
The populist pitch is pretty simple: the tech elite are a modern-day Marie Antoinette and their promises of AI-driven abundance are akin to societal brioche.
However, today, an inner city elementary school student has access to a high-quality tutor in his or her hand. 24 months ago, that would have cost them $50 per hour. Individual healthcare spending in America has crossed $15 thousand per year. Today, there are startups offering primary care for free thanks to collapsing cost of care with AI. And, while big city lawyers measure their hourly rates in four digits, a small business owner can get a redlining on their commercial lease for effectively nothing. I could go on for hours. The tokens that measure AI output are driving the largest democratization of services we have ever seen. We must continue to build and finance that effort.
The other end of the spectrum is just as interesting. As AI frees up time from menial tasks, the upper middle class and wealthy have an attention surplus to spend. Live events are exploding. A record 159 million people attended a Live Nation show in 2025. There has never been a generation that cares more about the quality of their food, clothes, and daily products. Pepsi spent $2 billion on Poppi last year to expand from sugar water to healthy sugar water. And as content creation gets commoditized, the price and value of premium media has gone up. Paramount is spending $110 billion on Warner Bros. Discovery for that very reason. It is the middle, the cheap and “value”-oriented, that is in secular decline.
Temporarily embarrassed millionaires are what make America so great; abundant intelligence gives anyone who wants to, the ability to drop those modifiers.
No free lunch
“Never was anything great achieved without danger" - Machiavelli
For the last fifteen years, and certainly the entirety of my career, balance sheet light was cool. In fact, not doing anything hard was cool. BaaS, embedded lending, and digital-native insurance agencies were all a bet on the thesis that venture dollars were meant to build technology, and technology only, and that real risk should be outsourced.
We got really pretty UIs masking origination businesses. And, it worked. Zero interest rate policy meant that capital was chasing any convexity it could find at a time where high-yield rates were under 5%. There were plenty of dollars from the likes of public pension funds that needed to meet their 7% hurdles and thus had to fund higher risk deals in pursuit of higher returns.
Then, all of a sudden, that ZIRP subsidy was replaced with 5% yields on 10-year Treasuries.
Consumer credit, the backbone of the US economy, is feeling that effect. 90 million Americans, or about a third of all US adults, currently use buy-now pay-later solutions. A quarter of them will use it for groceries. Yet, that pool of debt is not fully visible to lenders as Affirm and Klarna report some, but not all, of their products to credit bureaus and are hesitant to do so. With capital more expensive and data on consumers more opaque, consumer fintech startups no longer get to spend all their time on one-click origination. Bespoke underwriting, capital structure optimization, and debt service automation are cool again, or maybe for the first time.
Then, there’s the vanishing compliance and regulatory arb. The re-rate started with the Banking-as-a-Service middlemen who thought it made sense to rent someone else’s compliance and balance sheet so these so-called neo-banks could prioritize building tolerable interfaces for their clients. Synapse, one of these middlemen, had assets at partner banks for 18 million active customers until, in April 2024, their reconciliation process with one partner bank, Evolve Bank & Trust, broke and suddenly $265 million in funds were frozen. The rush to outsource difficulty and risk led to the liquidation of a $9 billion institution.
Just two years later, Column, a BaaS platform that actually went through the difficulty of getting a bank charter, has crossed $100 million of free cashflow. 20+ fintech startups from Revolut to Mercury have now filed for bank charters themselves. The era of rented complexity is over.
And, amidst all the chaos, risk is being re-priced everywhere and our insurance industry is seeing the consequences. Homeowners’ insurance carriers are literally withdrawing from states where they can’t price climate risks. State Farm, which had already not renewed 1,600 policies in Pacific Palisades, lost $7.6 billion on the LA fires. The war in Iran led to Lloyd’s syndicates backing out of war risk coverage in the Persian Gulf. The US government had to step in and guarantee maritime trade insurance through the US Development Finance Corporation.
At the same time that software gets commoditized, the plumbing that makes the world function is faltering. The opportunity to use technology to finance, bank, and insure assets has never been more compelling. But the value will accrue to those who put real capital behind their convictions. Returns are a function of risk. They always have been. There is no free lunch.
A vision for the next year
“The torment of precautions often exceeds the dangers to be avoided” - Napoleon
If you’ve made it this far in this letter, 2,777 words to be exact, you were either forced or are insane but, either ways, let’s close this out. Some of what drove me to write this letter is a deep frustration in the world I see around me, particularly in venture capital and the financing and building of new enterprises. We have become an industry of precautions.
The beauty of the SaaS business model corrupted our minds with easy money and it shaped the way we invest and what we invest in. The fight to minimize variance and chase certainty led to outsourced conviction, reverence of spreadsheets, and a mass influx of tourist technologists. We got extremely good at pattern matching to the last cycle with public comps and precedent transactions while forgetting that each new cycle in technology is a paradigm shift from the past.
We started chasing mediocre because mediocre is reliable and reliable gets through IC. Overdone financial models gave us something to blame when our fundamental convictions were wrong. And we convinced ourselves that price discipline was more important than exit discipline. Most firms are structurally incapable of betting on the future.
So who wins in this next era? When intelligence is abundant, difficulty is scarce. The victors will be those who do the things that most people can’t, or perhaps more importantly, won’t because they are “too risky” or “too complex”. Building physical infrastructure, delivering energy, and defending our way of life are hard and rewarding. Providing quality consumer experiences across healthcare, consumption, and commerce are both fleeting and valuable. And developing our financial system to serve our increasingly chaotic world is a crucial and profitable endeavor.
I’m more excited than I have ever been about what the future of technology holds. Here’s to the few outlaws building it!
- Atul Ajoy