State of Markets

Tech has gone from a “cottage industry” to the prime mover in the global economy. At this point, every company is, in some sense, a tech company–you’d be hard-pressed to find a truly pen n’ paper business.

More than any other sector, tech is driving investment, R&D cycles, earnings growth and margin expansion. The largest, most profitable, and fastest-growing companies the world has ever known, are tech companies. The AI infrastructure buildout alone is responsible for all or nearly all of net-new construction spending, blue collar job-growth, durable goods imports, net-new IG-issuance, etc. AI infra is arguably the only pro-cyclical impulse firing in the economy right now.

The point is that tech has left the cottage. Tech is the Everything Cycle now, and it’s not looking back.

The World’s Largest Companies Are Tech Companies

Tech Drives Value Creation & Earnings Growth

Tech Cos are driving the lion’s share of value creation and earnings growth, and have been for at least a decade.

Tech Drives Margin Expansion

Margins are historically wide, and while nearly every sector has seen margin expansion recently, tech has been the standout.

. . . Leading To (Very) Strong Fundamental Performance

Tech (and Energy) are leading earnings growth, but growth is broad-based: 94% of S&P500 companies met or beat EPS estimates.

Hard To Call It A “Bubble”

The market keeps setting new ATHs, but it’s profits, not prices doing the work. Despite growing profits (off an already high base), multiples have actually compressed, while performance is driven entirely by earnings growth

While Earnings Grow–off an already high base–multiples continue to compress

Multiple Compression Across the Board, But Especially Tech

Tech is up by more than 20% on the year, even with multiples ~20% below their 5-year average

The Market Is Not a Mag-Whatever Story, Anymore

Last year, performance and returns was a concentrated story: the Mag 7 dominated the field.

This year, the field has raced ahead, while the biggest tech companies have lagged the indexes. The most straightforward explanation for the shift is unpacked in the next section: hyperscalers are pouring their profits into the AI buildout.

Hyperscaler spending has been driving semiconductor earnings to unprecedented heights, and market returns have followed the money.

Momentum, Retail, and Leverage

Retail continues to be an important mover-of-markets, where leverage and momentum contribute to increased volatility. Unlike the meme-stonk era, retail has piled into the consensus themes.

We’ll Let You In On A Little Secret: AI Capex is Big. Really Big.

We Keep Underestimating The Capex To Meet Demand

Every year the street tries to forecast capex, and every year that forecast has fallen short. Demand for compute, and to some extent, prices, has lifted investment into Jackie Wilson territory: higher and higher.

Capex Is Eating Hyperscaler Cashflows (For Now)

Surging profits for hyperscalers have funded the lion’s share of capex, the flipside of which has been a steep drop in FCF that is expected to persist until ~2028.

Debt Markets Step In (On and Off Balance Sheet)

With hyperscaler cash fully-tapped, credit markets have sprung to life. Lending to Uncle Sam is no longer the only game in town.

Hyperscaler ROE Still Higher Than Their Cost of Borrowing

Debt certainly raises the stakes, and the numbers are big, but what ultimately matters is the return on investment. For now, hyperscaler ROIC is well-above their cost of borrowing.

These are (again) historically profitable companies, which until recently, borrowed very little. They are borrowing to invest in hard assets that are generating steady (and rising) cashflows. It’s a combination that lenders know and love.

Capex Is Having A Positive Impact On The Broader Economy

The AI buildout has led to a run on skilled labor, with a big premium to hard hat wages.

Surprise! Data Centers May Be Lowering Your Electricity Bill

Electricity costs, too, have come down due to economies of scale: a large, steady source of electricity demand covers an outsized share of the grid’s fixed costs, easing the burden on ratepayers.

"We estimate for every 10% increase in data center capacity average residential retail prices fell by approximately 0.4% on average . . . The result is interpreted thus: from 2019 to 2024 the average residential customer lived in a state where data center capacity grew 160%, which caused their rates to fall by 6%." Have Data Centers Raised Your Electric Bill? Watten, Bistline, Blanford (Aug. 24, 2026)

AI Is Showing Up Everywhere

Revenue Growth Is Unprecedented For Speed and Scale

The justification for all this investment is straightforward: demand for AI is like nothing we’ve ever seen.

Hyperscalers Keep Saying The Same Thing: Demand > Supply

Cloud revenue backlogs have doubled yoy, and while cashflow is near-negative for now, it’s expected to come roaring back in two-years’ time.

Neoclouds Are Also Growing Historically Quickly

Reports Of GPU Obsolescence Are Greatly Exaggerated

Every new generation of chips was supposed to render the older generations obsolete, the bears claimed. “The depreciation schedules keep getting longer–what is this chicanery?!” they howled.

It turns out that depreciation schedules might not be long enough. As demand for compute keeps accelerating, GPU rental prices and residual values keep climbing and/or holding steady, even for the oldest generation of chips.

It’s not chicanery, so much as simple unit economics, that have made all this capital allocation look pretty smart, thus far.

Despite All The Demand, It’s Still So Early

AI adoption is broad and impactful, but still relatively shallow.

Anecdata Around AI ROI Abound

Amazon

AI Shopping & Ads “...customers who use Alexa for shopping spend an average of over 40% more per order…” “Ads Agent…6% lower cost-per-acquisition…” Earnings Release, Amazon, Q2, July

Meta

AI Ad Performance “...GEM… a 15% uplift in conversations on Facebook” “...Advantage+… 13% incremental lift in purchases and a 16% increase in add-to-cart conversions.” Susan Li, CFO, Meta, Q2, July 2026

Shopify

AI Ad Performance “...Sidekick…drove like an 8% increase in new merchants reaching 5 orders within 15 days.” “AI search…80% higher than traditional organic search.” Harley Finkelstein, President, Shopify, Q2, August 2026

ServiceNow

AI ACV & Deployments “ServiceNow AI crossed $1 billion in annual contract value in Q2 2026” Earnings release “...Agentic deployments of ServiceNow AI increased ninefold in just 9 months.” Bill McDermott, Chairman and CEO, ServiceNow, Q2, July 2026

Affirm

AI Underwriting Affirm used the model to approve. . . applications that its existing system would have declined, including . . . limited credit histories and no FICO scores. . . that produced 3.4% more completed purchases, and those additional loans performed better than a comparable expansion." Company release Sept. 2026

Lemonade

AI Claims Handling “Our LAE ratio declined to a record low 5%... continued advances in AI and automation across our claims operation…” Shareholder letter, Lemonade, Q2, July 2026

Chime

AI Cost Efficiency “...AI also continued to drive outsized efficiency gains…reduced our cost to serve…10% for each of the last four years.” Chris Britt, CEO, Chime, Q2, August 2026

Snowflake

CoCo Adoption “CoCo surpassed 9,100 accounts… adding more than 2,000 accounts in the quarter alone…” Earnings release, Snowflake, Q2 FY27, September 2026

Microsoft

Copilot Revenue & Seats “...GitHub Copilot… Copilot revenue accelerated over 60% quarter-over-quarter.” “...We now have over 30 million paid Microsoft 365 Copilot Seats…” Satya Nadella, Chairman and CEO, Microsoft, Q4 FY26, July 2026

Walmart

AI Shopping "The number of customers using Sparky is up 70% from last year…” “...Customers and members who use Sparky for shopping spend 40% more per order than other who don’t.” John Furner, President and CEO, Walmart, Q2 FY27, August 2026

Enterprise AI Spending Has Plenty of Room-to-Run

Most companies are not cost-constrained, and intend to spend more on AI.

In the big scheme of things, AI spend is a tiny share of the overall cost stack.

Consider that cloud reached ~4% of IT spend in year two. Most estimates of AI spend are below that, and TAM for AI is considerably larger than IT budgets.

Let’s Just Say That There Were Signs

Elsewhere, there’s been an explosion of code, and a surge in entrepreneurship.

AI Intensity Leads to MORE Hiring, Especially For SWEs

More Signs That We’re Still Early: Beneath the Surface, AI Is Not Normal

The impressive aggregate picture actually obscures what is going on beneath the surface: adoption is wide, but mostly shallow (for now).

Growth is driven by a tiny share of power-users, who are unlocking more value with increasingly sophisticated tools and implementations.

As the cost of intelligence falls, demand continues to increase for both frontier and non-frontier models. Agents especially are driving parabolic growth in token demand, as routing, caching, and other efficiency gains, make them economically viable. It’s Jevons Paradox in real-time.

The exponential increase in demand is driving the price of compute upwards–even 4-5 yo GPUs cost roughly the same as they did a year ago.

And all of this is driven by just the top 1% of users, who are only beginning to scratch the surface of what’s possible. Eventually, the rest of the market will catch on (or else be left behind).

Power Users Are Pulling Away

A tiny fraction are using orders of magnitude more AI than the field, and the gap has been widening since early this year.

Jevons Paradox: Cheaper Intelligence, More Demand

Cheaper Tokens Have Ushered In The Age of Agents

. . . and Token Demand Goes (Even More) Parabolic

Instead of Open v. Closed, Why Not Both?

Cheaper intelligence is expanding the surface area of demand, which means that model competition is not a zero-sum-game.

To the contrary, tiered performance and pricing is giving firms more opportunities to unlock use-cases, up and down the value chain.

In the bigger scheme of things, open-weight tokens are increasing share, but overall token spend is increasing, as well, with frontier models still capturing a premium for frontier problems.

As Users Optimize Their Spend, Demand for Compute Only Builds

Like GPUs, older models are far from obsolete. As the overall cost of intelligence falls, the pie keeps growing, with frontier and non-frontier models splitting share.

Increased efficiency begets more demand, driving compute prices steadily upwards.

On the Consumer Side, Things Are (Also) Just Getting Started

Consumer AI: Search Referrals Gaining (but still very small)

Waiting for the SuperApp: Muse Has Entered the Chat

Not The First Capex Rodeo: Semis Led Software Before

AI has led to a broader rotation from bits to atoms

Follow The Money

Returning to markets, all of this AI supply and demand, has had a profound impact on the types of businesses that are driving performance. For more than a decade, software was the star of the show, but with the advent of AI, hardware has taken center stage.

The multi-trillion capex super-cycle has caused a surge in demand for, among other things, GPUs, HBM memory, power components, skilled trades, and even copper, that has far outstripped supply. The result has been historic earnings growth for traditionally sleepy, cyclical sectors, and stock prices have responded, accordingly.

To be clear, it’s not only AI-infra driving the rotation from Bits to Atoms. EVs and solar power have also precipitated a massive grid glow-up. Defense spending is on the rise. Rockets and robots are both emerging new frontiers. Global politics and secular aging have lent urgency to the task of rebuilding an industrial base that will support core manufacturing, innovation, and automation.

We’ve entered a new cycle of real assets, and while tech is running through all of it, capital-light bits is no longer the (only) name of the game. Another Age of Atoms has arrived.

Follow The Money: Capex → AI infra (cont.)

The Rotation From Bits to Atoms Is More Than Just Chips

Infrastructure and industrial needs are massive. Globally, ~$90T of investment is necessary by 2040.

A Growing Manufacturing Base

Electrification of Everything

Rockets To More Than Just The Moon

We’re Going To Need More Robots, And Those Robots Need More Parts

AI is triggering another great wave of automation and robotics. As of now, most of the supply chain is overseas. That’s changing.

Robotaxis Inbound

Rising Defense Spending

The Rotation To Atoms Has Flipped The Field In Markets

Unsurprisingly, this great leap forward in Atoms investment is showing up in market pricing. After a decade-long of capital-light dominance, capital-heavy businesses have flipped-the-field.

It’s not that capital-light businesses (like software) have performed poorly (SaaSpocalypse notwithstanding), so much as tech’s largest and most profitable companies–and thereby the world’s largest and most profitable companies–are pouring those profits into hardware like never before.

The net result is the AI Supply Chain has been the winning trade, thus far. That’s the chipmakers, the optical interconnectors, the turbine shippers, the liquid coolers, and everything in between.

The AI Supply Chain Has Been The Winning Trade

. . . and there’s been no bigger winner than memory.

What If Semis Are Still Cheap?

The funny thing is, though, that as stellar as market performance has been, there’s an argument that it hasn’t been stellar enough. Forward earnings multiples for some of the largest chipmakers, are well-below what the sky high expectations for earnings growth would normally predict.

There too the reason is fairly straightforward: demand is highly contingent on hyperscaler capex, and seasoned investors in cyclicals are always wary of the cycle’s peak.

It’s also true that necessity is the mother of invention, and with great profit margins, one can expect great efficiency gains for power, compute and memory, right around the corner.

But capex shows no indication of a slowdown any time soon, and if demand for intelligence is indeed infinite, then this cycle may not be like the other cycles.

If that’s the case, then future investors are going to have a hard time looking back on those multiples . . . between all the tears, that is.

The Semi Supply Chain Remains Backlogged

Whatever skepticism may preside, ~3 years into the buildout, supply is still lagging demand.

Somehow the GPUs keep arriving on time, but every other piece of the supply chain continues to have unusually long lead-times.

And even if you can get your chips, it’s an open question if you can power them up.

More Like Software’s Selective Selloff, As SaaS Enters a ‘Prove-It’ Era

More Prove-It Than Pocalypse

The flipside of hardware’s outperformance, has been software’s relative malaise. 2026 began with the SaaSpocalypse: a seemingly undifferentiated sell-off of formally high-flying public SaaScos, as AI prepared to eat their collective lunch.

What’s transpired, however, is better described as “prove it,” than -pocalypse.

SaaS has re-rated (gradually, and then suddenly) to reflect the industry’s broader shift from growth to profitability. As the share of high-growth, less-profitable companies has given way to slower-growth, more-profitable companies, multiples have compressed accordingly. At the same time, if you’re still a high-flyer, your multiples are still pretty, pretty good.

Likewise, AI has certainly raised some doubts about defensibility and terminal value, but not uniformly. Companies with perceived (and/or demonstrated) defensibility and/or AI tailwinds, have outperformed. Cyber and observability has been a standout category, and vertical saas has experienced less compression than horizontal.

Regardless of market-pricing, and the disruption that AI may or may not bring, it remains the case that the underlying fundamentals for SaaS has been remarkably stable thus far. Both revenue growth and operating leverage have showed little signs of deterioration (yet), even if their not quite as impressive as before.

Software’s Selective Selloff: Cyber > Vertical > Horizontal

Software Multiples Compress, But Not Uniformly

Growth And Operating Leverage Stable Or Improving (Mostly)

In The Bigger Scheme Of Things, SaaS (And Software) Have Traded Growth For Profitability

The share of high-growth software companies is a fraction of what it once was, but profitability is now the norm.

Will Software Go the Way of Print Media?

. . . Or Is AI A Rising Tide of ROE to Lift All Boats?

An Historically Large IPO

The Most Valuable Private Cos Are Larger Than Ever

. . . And value creation is happening on the private side.

Private-For-Longer: Skip Small Cap, Go Straight to Mega

Power Law Abides, But Even More So

Secondary Markets Tell The Same Story

Discounts are few and far between, and the best companies trade a premium

Power Law Applies To Managers, As Well

For VC Funds, The Difference Between Top Decile And The Rest Has Never Been Wider

Adjusting To Life After ZIRP

Not unlike Public Cos, when ZIRP ended, startups traded growth for profitability.

That said, raising fresh equity tends to require rapid growth. The companies who have successfully raised recently are still growing at a pace in-line (or faster) with peak-ZIRP.

Everything Is AI Computer

This cycle is not like the previous ones

Concentration has been a theme of the new cycle. A smaller number of funds are deploying at larger sizes into fewer startups in, an increasing share of which are located in Silicon Valley, and nearly all of them are “AI-related.”

There is one sense, however, in which all that concentration is a bit misleading. Beneath the surface, there has been a mix-shift in the kinds of companies that are being funded. Venture dollars are flowing to enterprise software and consumer apps, as they always have, but also traditionally out-of-scope services (e.g. legal, accounting) and formally less-popular capital-intensive verticals, like semis, power, and defense.

A New Breed Of Unicorn

Bigger, faster, younger.

The AI Generation Is Built Differently?

One under-considered recent phenomenon is the extent to which private techcos are increasingly generating the leading signal on the most important economic questions (and often telling their own stories in the process)

Token Rails

Openrouter has become a leading source of data to observe the parabolic growth of token demand

AI Adoption Is Wide and Shallow (For Now)

OAI research and data have highlighted the not-normal distribution of AI usage

The Token Efficient Frontier

Companies like Databricks have published research demonstrating model optimization strategies, as cheaper intelligence brings more use cases into scope

Pricing the Forward Compute Curve?

The price of compute has become on the most important KPI behind the trillion dollar buildout, and Kalshi has been making the market

Inference & Insinuations

Outside of the leading clouds and neoclouds, a new infrastructure layer is emerging. Specialized inference providers provide visibility into the speed and scale of the ecosystem’s growth.

SaaS . . .acceleration?

Stripe has unmatched visibility into payment flows, and their data is even more counter to the SaasPocalypse narrative: far from dead, SaaS revenues accelerated into the new year.

AI is Saving Lives

That AI will revolutionize even the most labor-intensive services is our story, and we’re sticking to it. It’s early, but Protege data shows some early returns: AI is already saving lives

Making life easier for renters and owners with AI

AI is diffusing into the broader economy in disparate ways. Sometimes the benefits are substantial, but often they are incremental at the margins (which, nonetheless, have a tendency to add up to something quite substantial).

In late November 2025, EliseAI worked with Elevenlabs and Cartesia . . . to collect a series of high level voices with specific regional accents curated from their robust voice library . . . The underlying hypothesis was that, given existing research showing consumers are more likely to transact with people who sound like them, that switching to localized voices would positively impact lead-to-tour conversion rates.

A New Defense Prime

Not only is defense spending rising, but priorities have changed. Munitions are smaller, cheaper, and built for scale.

Stopping Crime Is Good Actually

Hardware innovation is making a difference beyond chips and defense. ALPRs are a force-multiplier for chronically understaffed police departments.

Collectibles Are Timeless

Amidst all the AI, there are other changes afoot. Social media has upended traditional retail distribution, and creators are building their own brands. Combined with an increasing desire for scarce things (like collectibles), it’s led the birth of a new QVC.

• Heavy-use consumer
• Robotics
• Autonomy
• AI x Bio
• Personal Health
• Diffusion into the Enterprise beyond Coding
• New era of American Dynamism