Let It Ride

At the end of 2025, about 33% of U.S. household wealth was in stocks. This beats the 30% during the meme craze in 2021 and the 27% during the height of the dot.com bubble in the first quarter of 2000. According to the BEA’s Q1 2026 GDP estimate, computer and peripheral equipment investment grew at a 67.4% annualized rate. Strip out AI-driven categories, and headline GDP growth would fall from 1.6% to 1.0%. If you strip out AI-adjacent investment activity such as the construction of data centres, GDP growth might vanish altogether. This partially explains why many Americans feel the economy is poor despite an ebullient stock market.

I don’t know if we are in a bubble or not. If super-intelligence is around the corner and a future AI will be like Johnny Depp in the movie Transcendence, then maybe the market is cheap. Hock Tan, the CEO of Broadcom, said “We’re all learning. I haven’t figured it out yet... A year from now, what I think I know today might yet be proven wrong… We are all running a full-scale experiment, all of us, in broad daylight, spending hundreds of billions of dollars a year. That’s what we are doing. It’s kind of fascinating.”

Leaving aside AI for the moment, I’ve invested through the dot.com era, the global financial crisis and the 2021 meme/crypto/SPAC bubble and ultimately those bubbles were created by too much money.

The dot.com bubble was fuelled by Alan Greenspan’s 75-bps cut for the Long-Term Capital Management crisis in late 1998, and fear of Y2K. He used monetary policy to help smooth markets sending the risk-free rate well below the inflation rate. The result was cheap money and an epic run in dot.com stocks. Ultimately this bubble was popped as interest rate hikes tightened the money supply and valuations compressed to multiples of cash flow rather than multiples of hope. The fun was over and the Nasdaq fell 78% from its peak.

The GFC was fuelled by credit creation, another way to create money. Investment banks leveraged their own balance sheets for investments whether in structured credit or anything else they wanted to buy. They helped create the sub-prime bubble where anyone could be handed the keys to a house with a bad credit history. When the impairments started, there was a sudden contraction in the money supply from bank deleveraging, the financial system started breaking down and the Fed was forced to print money and save the banks.

The 2021 bubble was money creation from the fiscal side. The $5 trillion+ in pandemic-era fiscal stimulus, culminating in Joe Biden’s $1.9 trillion American Rescue Plan fuelled asset prices. SPACs with no revenues and dreams could get funding. This bubble popped as the Federal Reserve rapidly lifted rates due to high inflation.

In summary, past bubbles were fuelled by monetary easing, credit creation and fiscal deficits. Looking at today’s conditions, we have all three at once.

First, the Fed started to lower rates last year, with the White House saying it’s not enough. Rates are currently below the inflation rate like the dot.com bubble.

Second, bank balance sheets have increased thanks to the Fed reducing the enhanced SLR requirement for big banks from 5% for bank holding companies and 6% for bank subsidiaries down to a range of 3.5%-4.5%. This relief directly frees balance-sheet capacity for banks to hold more Treasuries and extend more repo financing without it counting against their capital ratio. The Treasury will issue $2 trillion of new bonds this year and needs someone to buy them. The banks are a natural counterparty, but they were complaining to the Fed that these Treasuries were using up valuable balance sheet. This change allows banks to hold more Treasuries and do their other lending.

This is also the mechanism that allows banks to give virtually unlimited margin to hedge funds. Because Treasuries are considered an ultra-safe asset, bank margin requirements can vary from 0% to 2%. Gross hedge fund borrowing hit a record $6.2 trillion in Q1 2025, up 26% year on year, with average gross leverage at a five-year high of 294%. Hedge funds buy Treasuries from banks, borrow from banks and post little equity. The hedge funds then sell the Treasury futures, which usually trade at a slight premium to the cash market. Net Treasury inventories at primary dealers have risen to $550 billion in 2026 up from $400 billion in 2025, a 40% increase augmenting the trade and allowing the U.S. government to help finance its debt binge.

We’ve also seen credit creation through the private credit market. If Elon Musk isn’t recursively generating robots, the data centre could become the sub-prime crisis of our time.

The biggest money supplier of all is the U.S. federal government with a 6% budget deficit, the highest sustained peacetime level in history, exceeded only by the GFC and COVID. About $2 trillion of new money will be created this year.

All this money needs to go somewhere. In 2025, it was going to gold, or foreign currencies, partially to stocks. But now it’s all going to AI labs, SpaceX and whatever else is necessary to make our lives a utopia or dystopia, depending on your world view.

The most likely lever that would pop the bubble in the near-term is interest rates (if it’s indeed a bubble, like Hock Tan, I don’t know; maybe the massive bet works). Which brings us to the Middle East.

Iran and America Are About to Sign a Deal (Though This Round Comes with Higher Stakes Than the Previous 38 Times). It’s hard to know for sure how close a deal is because apparently, they only speak to Iran’s new Supreme Leader through human couriers (are carrier pigeons too risky?). This is operational security for a leader under active assassination threat.

Bloomberg sources among diplomats intermediating talks suggest there is an 80%-85% chance of a deal. Iran’s leader can’t do a social media post, so we’ll have to wait. Several hardliners want to hold out because they think severe global inflation will make Trump offer more concessions.

Without a deal, serious inflation looks all but certain. The Straits of Hormuz have been closed and Shell CEO Wael Sawan said in an interview this week that the global oil market is short 1.2 billion barrels of crude. Inflation would force the Fed to raise rates, a scenario that could impact equities, gold, private credit and hedge fund financing all at once. Scott Bessent at the Treasury Department will find fewer buyers of Treasury bonds, which could widen the term premium or (horrors) force the government to cut spending.

DPI is the New IRR. Apollo’s Co-President Scott Kleinman said at SuperReturn Berlin that private equity firms that deployed capital in 2017-2022 lost their way and overpaid for assets in a zero-interest rate environment when “valuations always went up” and now face a reckoning as rates normalize.

This is why DPI (Distributions to Paid-In Capital) has become the metric that matters. The hold time for private equity assets has doubled from a historic average of around four years to almost eight years today, leaving a $4 trillion overhang of assets waiting to be sold as sponsors face pressure to return capital to investors. Investors now want to see the cash. 74% of institutional LPs now rank realized distributions as their primary evaluation criterion. And they are right to. LPs are receiving roughly half the realized cash they received in the late 2010s while paying the same management fees on the same NAV.

As DPI gains ascendance, valuation capitulation is inevitable.

SaaS Incumbents Lose Pricing Power. Adobe reported earnings and showed intent-based discovery is replacing traditional software adoption. Users search for outcomes like “summarize this PDF” rather than buy products. This forced Adobe to provide immediate gratification without paywalls. Despite 40% traffic growth in its AI users, Adobe found its traditional sales funnels weren’t converting the AI-driven traffic as much as they would like. Adobe has focused on freemium offerings to expand its user base and integrate with AI platforms rather than trying to wall off its ecosystem.

Adobe is afraid of AI-native competitors eating its funnel. It’s unclear whether the high growth rates in AI-first ARR will compensate.

Another small datapoint. We were looking at a CRM system for our marketing campaigns. The current one is working well, but the vendor wanted to aggressively price hike, so we decided to look at options. Another hedge fund COO had a similar issue and used Claude Code to make a CRM system. He spent a week working on it, and now the firm has an internally made CRM system. He offered to share the code on GitHub. While it’s unclear whether we will like his system, we can fork his code or make the code from first principles ourselves. This frees us from vendor lock-in and having to spend the time on renewal contracts in the future.

Signs of Excess. Goldman put out a piece this week raising its 2027 hyperscaler capex bull case to a whopping $1.4 trillion. Consensus capex spending is $920 billion for 2027. They base its number on historical infrastructure cycles noting that railroads and autos reached 2%-3% of GDP versus AI’s current 1.5% of GDP. This implies that AI could still have room to grow. Some anecdotal evidence that Goldman points to on how spend could scale is that Google Cloud and AWS backlogs grew from $358 billion to $832 billion in six months. Goldman also projects 24x token consumption growth through 2030 which could further grow capex.

To put these numbers in context, global spending on cancer treatments is $290 billion. These figures are truly gargantuan.

Fortune Brainstorm Tech had a panel of CTOs and AI executives diagnose why AI ROI is elusive. Their view is the bottleneck isn’t capability, but organizational and architectural readiness. State Street’s CTO said the chokepoint was that the data needs to be in the right places with the right controls before AI deployment. In regulated industries, this is a multi-year remediation project. Deloitte’s Vice Chairman of TMT said the fatal mistake is dropping AI into legacy workflows. Existing inefficiencies get “weaponized at scale.” Real productivity requires redesigning processes from scratch, which is far harder and slower than automation. Lambda’s CTO said outside software development agents aren’t yet fit for purpose in most enterprise domains. Wells Fargo’s Chief AI Product Officer noted that even when AI works, quantifying returns is non-trivial. A 25% uplift in account openings is measurable. The value of deeper client relationships is not. This creates internal budgeting friction that slows rollout.

This signals the AI ramp may be more long dated than what the market is pricing in now.

End Note

$109,700 is now the ‘low income’ threshold in San Francisco for a single person. With SpaceX cafeteria workers becoming millionaires, I suspect this threshold will push higher.

Apple unveiled Siri for the AI age. Their presentation sounds almost exactly like the one they made when they launched the iPhone 4S in late 2011. Sometimes as much as things change, they stay the same.

Omar Sayed