Paladin October 2026 Economic Outlook

Executive summary

The U.S. expansion rests on two narrow supports, AI capital spending and consumption by the wealthiest households, and I believe both depend on confidence in AI economics that the numbers don't yet justify. I expect this to resolve within about two years: either AI revenue shows up across the broader economy, or the gap becomes obvious.

The four largest hyperscalers are on pace to spend roughly $730B on capex in 2026, against about $115B in combined annualized revenue at OpenAI and Anthropic. One year of spending runs about six times the revenue it is supposed to earn. If financing for the leveraged part of that buildout tightens, I expect a capex pullback, an equity repricing, and a sharp retreat in top-decile spending. That combination produces a recession.

Rates would fall, the front end first and the 10-year less. The catch is the debt math: in every recession since 1980, nominal growth has fallen below the 10-year yield, which pushes debt-to-GDP up just as the fiscal cushion is nearly gone. With debt held by the public near 101% of GDP, I expect financial repression and higher taxes to follow.

For real estate, the central case is an opportunity for those positioned for it. A downturn halts new construction and forces sales by owners with floating-rate or maturing debt, which opens the door to buying below replacement cost. At Paladin Equity we are preparing with three plays: value-add multifamily, hospitality with a focus on extended stay, and tax strategy, while making sure the portfolio survives near-term strength or stagflation.

Part 1: The two engines of growth

AI investment and top-decile consumption are carrying an outsized share of U.S. growth, and both depend on the same thing: confidence in AI.

AI capex. ING estimates the tech investment rollout accounts for about a third of current year-over-year U.S. GDP growth (ING). Its broadest measure put tech investment at 50.2% of growth in Q2 2026.

Top-decile consumption. Moody's Analytics estimates the top 10% of earners drove 49.2% of consumer spending in Q2 2025, the highest in data back to 1989 and up from about 36% three decades ago (Crain Currency). Moody's chief economist Mark Zandi has warned that if these households turn cautious, the economy will slide into recession.

The two engines are linked. Equity gains in AI names fuel the wealth of top-decile households, and that wealth fuels their spending. A hit to one is a hit to the other.

The rest of the economy is stagnating rather than contracting. Spending by the lower third of cardholders shrank in mid-2025 and stayed nearly flat into early 2026 (Minneapolis Fed). That segment cannot absorb a shock, so growth rests on the two engines.

Part 2: The capex-to-revenue gap

One year of hyperscaler capex is roughly six times the combined annualized revenue of the two leading AI labs.

Amazon, Alphabet, Microsoft and Meta plan about $730B of 2026 capex, up roughly 78% from about $410B in 2025 (ValueAdd VC tracker). CreditSights projects capex at about 86% of sales for Oracle, 54% for Meta, 47% for Microsoft, 46% for Alphabet and 25% for Amazon (CreditSights).

Two caveats make that revenue look stronger than it is:

• Accounting. Anthropic counts sales routed through AWS and Google Cloud; OpenAI does not (BigGo / FT). OpenAI's widely reported ~$68B figure was corrected to ~$50B in October, and Nvidia, Oracle and CoreWeave fell on the news (Quartz).
• Tokenmaxxing. Much of the 2026 surge came while companies pushed employees to maximize AI usage, running up token bills with unclear payoff (Let's Data Science). Some of that usage gets cut once CFOs audit it.

Pricing pressure is already showing. In September, OpenAI passed Anthropic in OpenRouter spend share for the first time in 2.5 years (OfficeChai), and Chinese open-weight models now carry a large share of token volume at a fraction of the price. OpenRouter is a small window on the market, but the direction is clear: revenue per token is under pressure.

Part 3: AI erodes the cash flows funding the buildout

The hyperscalers fund most capex from operating cash flow, but the same technology threatens the franchises producing that cash.

Search and advertising. Alphabet's search and Meta's advertising businesses pay for a large share of the buildout. AI answers reduce clicks, shrink ad inventory and weaken the link between a query and a paid result. Unless Alphabet earns search-level margins on AI answers, its core cash engine erodes even while the company stays diversified.

Software. AI's biggest gains so far have come in coding, and coding is aimed straight at software companies. In early October, developer Brandon Thomas released free, open-source rebuilds of seven Adobe apps, including Photoshop, Premiere and Lightroom, written with AI coding tools (TweakTown; Notebookcheck). The apps are early alpha builds and not ready for professional work. But the cost of copying a professional software suite has collapsed, and every high-margin subscription software company now faces pressure on its free cash flow.

Value destruction without value capture. AI moves value out of incumbent profit pools and hands it to consumers, often free or below cost. The labs don't capture it at the same margins, and the incumbents lose it. That's how AI can disrupt existing revenue and still fall short of paying for its own buildout.

Timing is the counterpoint. So far AI has improved ad targeting, and search revenue has held up, so the erosion is moving slower than the capex cycle. Expect it first in thinner margins and lower valuation multiples, and only later in revenue.

Part 4: The leveraged layer breaks first

If this cracks, it likely starts where expensive debt meets unproven revenue: Oracle, neoclouds such as CoreWeave, private-credit data center vehicles, and the labs that depend on outside capital.

OpenAI is the clearest example. It has signed up to $1.4 trillion in compute commitments over eight years (Fortune). Per the Financial Times, it projects about $278B of cumulative negative free cash flow from 2026 to 2030, with the $122B raised in March exhausted by 2028 (Sacra). It posted a net loss of $38.5B in 2025 on $13.07B of revenue (Yahoo Finance).

The deals are circular. Chip and cloud vendors invest in labs that then buy their chips and capacity, and some backers fund their stakes with debt. SoftBank's $10B OpenAI investment is largely debt-financed (BigGo / FT). When one link needs fresh capital and can't get it on good terms, demand falls across the chain at once.

The likely trigger is a financing market closing on this layer, not a hyperscaler deciding to pull back. The hyperscalers then follow by trimming capex guidance.

Part 5: The wealth effect reverses

An AI-led equity repricing hits top-decile spending directly, because that spending has tracked these households' portfolios.

The typical top-10% stockholder owned $1.1M in equities in Q3 2025, up from $624,000 at the end of 2022 (AOL / Moneywise). Zandi estimates every $1 of added net worth eventually lifts spending by two cents, and that the wealth effect added a full point to consumer spending growth in 2024 (Fox Business). That channel runs in reverse when stocks fall.

AI names make up a large share of the major indexes, so an AI repricing is a market event, not a sector event. Capex and top-decile spending fall together, and the bottom 90% has no slack to make up the difference.

The size of the K is debated. The BLS Consumer Expenditure Survey puts the top 10% share of spending at about 23%, broadly stable since 2004 (Franklin Templeton). The true figure likely sits between a quarter and a half. Either way, high-income spending is a major and asset-sensitive driver of demand.

Part 6: Rates and the deficit

I expect a bull steepener: short rates fall quickly, and the 10-year follows, but less, to the degree inflation has cooled. Rate cuts ease the interest bill, but a recession widens the deficit far more than cuts save.

Why the Fed would cut fast. CBO projects a $1.9T deficit for fiscal 2026, with net interest at 3.3% of GDP and the primary deficit at 2.6% (CBO). Interest is more than half the deficit, and the Treasury paid $963B in net interest in the first ten months of the fiscal year (The Wealth Advisor). Bills reprice within months, so front-end cuts relieve the budget quickly.

Why the long end falls less. History supports lower long rates in recessions as money moves into Treasuries, but three forces limit the drop:

• The primary deficit widens. Tax receipts fall and automatic stabilizers kick in. In 2009 the deficit jumped from about 3% of GDP to about 10%.
• Maturity structure. Only about a fifth of marketable debt is bills, so most interest savings arrive gradually.
• Inflation. In a pure demand shock, long yields fall well. If oil, tariff or fiscal inflation is still running, they fall much less.

The starting point. Of $32.4T in marketable debt, $7.25T is bills, $16.2T notes and $5.5T bonds. About 33% matures within 12 months, and average maturity is 71 months (JEC). The weighted-average rate is 3.53%, with bills at 3.87% (Treasury data via PrimeRates).

What cuts save. Assume the Fed cuts 200bp, bill yields fall from about 3.9% to 1.9%, and 2- to 10-year yields fall about 100bp.

A sustained 100bp drop across the whole curve eventually saves about $320B a year, but only about a third shows up in year one. The Fed adds a lagged bonus: it pays interest on roughly $3T of bank reserves, so a 200bp cut saves it about $60B a year, which first pays down its accumulated losses before reaching the Treasury.

What a recession costs. A mild recession with only automatic stabilizers widens the primary deficit by about $500–700B (1.5–2% of GDP). A moderate one with stimulus adds about $1–1.5T (3–4.5% of GDP). For reference, 2001 swung the deficit about 3 points of GDP and 2009 about 7.

Cuts offset roughly a sixth to a quarter of the added deficit in year one. They shrink interest's share of the deficit, not the deficit itself. The payoff comes in years two and three, as the primary deficit narrows in recovery and interest savings keep compounding.

For borrowers, refinancing conditions would likely improve, but by less than T-bill rates suggest.

Part 7: The debt math and financial repression

Whether a country can grow out of its debt comes down to one comparison: is nominal growth running faster than the interest rate on the debt? That cushion has nearly closed, and a recession would push it negative.

The 10-year tracks trend growth. Across 1956–2025, the annual correlation between nominal GDP growth and the 10-year yield is 0.52. Against a five-year average of nominal growth it rises to 0.84. The 10-year prices where nominal growth is heading over years, so a slowdown in trend growth from about 5% toward 3% points the 10-year toward the 3–3.5% range over time.

The 1956–1980 regime shrank the postwar debt. The 1981–2007 regime made debt expensive. The 2008–2025 regime held yields down through QE and zero rates, which swamped the year-to-year link; against the five-year average of nominal growth, the 2008–2025 correlation is 0.65, and the annual figure is distorted by the 2020–2021 pandemic swing.

The cushion is nearly gone. Growth minus yield was +9.6 points in 2021, +6.8 in 2022, +1.4 in 2024 and +0.7 in 2025. In every recession since 1980 it turned sharply negative: −8.7 in 1982, −4.6 in 1991, −1.8 in 2001, −5.2 in 2009 and −1.7 in 2020. Yields fall, but growth falls further.

With debt held by the public near 101% of GDP, each year of yields above growth pushes the ratio up. An AI bust would deliver lower rates and worse debt math in the same window.

Financial repression. History shows how governments handle that combination. Financial repression means holding interest rates below inflation and nominal growth for years, while rules and incentives steer banks, pensions and savers into government debt. The debt shrinks relative to the economy, and savers quietly pay for it. That is how the U.S. worked down its World War II debt. I expect a recession to set that regime in motion, alongside higher taxes, because there is no painless way out of debt at 100% of GDP.

Part 8: Scenarios, signposts and the real estate opportunity

My central case is a downturn with falling rates within about two years, the window in which I think this resolves. The scenarios below are judgment calls, not model outputs.

The AI-revenue scenario requires revenue to show up across the broader economy, not just more hyperscaler spending. More capex on its own only delays the outcome and widens the gap. Stagflation is the hardest scenario for real estate: cap rates rise and refinancing gets harder while operating demand weakens.

What has pushed me toward the downturn case is the OpenAI revenue correction, the shift in OpenRouter spend share, and the tokenmaxxing evidence. Each year of $700B+ capex raises the stakes if revenue growth stalls.

Signposts that would confirm the thesis before the broad market prices it in:

• A hyperscaler cutting capex guidance, or declining to give a 2027 outlook
• OpenAI raising at a flat or lower valuation, or failing to close its next round
• Widening credit spreads on Oracle, CoreWeave and other neoclouds
• Falling GPU rental prices and idle capacity
• Slower growth in reported AI lab revenue, or further restatements
• Weakening top-decile spending in card data from Bank of America and others

Signposts that would argue against it: AI lab revenue that keeps compounding on a net basis, enterprise adoption that broadens well beyond coding tools, and hyperscalers showing rising AI-driven margins rather than just rising capex.

Why the central case favors real estate buyers. A downturn stops new construction. Lenders pull back, and the supply pipeline dries up for years. At the same time, owners with floating-rate debt or maturing loans are forced to sell. That is when good properties trade below replacement cost, meaning for less than it would cost to build them today.

• Hospitality, especially extended stay. Extended stay runs lean, has historically held occupancy better than traditional hotels in downturns, and serves workforce, relocation and insurance demand that doesn't go away.
• Multifamily value-add. People still need somewhere to live, new supply stops, and buying below replacement cost leaves room to renovate and raise rents.
• Real assets under financial repression. Real assets financed with long-term, fixed-rate debt are one of the few places that come out ahead when rates are held below inflation.

Part 9: How Paladin is positioning

We are preparing for the downturn on a roughly two-year view while making sure the portfolio survives near-term strength or stagflation. We are doing it with three plays.

1. Value-add multifamily. We are lining up capital and relationships now so we can buy below replacement cost when sellers are forced to move, then renovate and reposition. We underwrite with fixed-rate or hedged debt we could carry through stagflation, and we don't underwrite deals that only work if rates fall.
2. Hospitality, with a focus on extended stay. Through Paladin Hotels, we operate four Everhome Suites extended-stay properties in Southern California, so we know where the operating upside is. Distressed hotels with good bones are where discounts will be deepest, and the lowest-cost operators win when demand softens. We are also planning for weaker corporate and project-driven demand in a recession, particularly business travel and extended-stay segments tied to construction and tech.
3. Tax strategy. With deficits at these levels, financial repression probably won't be enough on its own, and taxes are likely to rise. Real estate remains one of the most tax-efficient ways to build wealth, through depreciation, cost segregation and 1031 exchanges. Structuring for after-tax returns matters more when tax rates are rising and real interest rates are held down.

On financing, expect relief to come mostly at the front end. Long-term fixed financing may improve less than headlines suggest. The best opportunities usually come 12–18 months after the break, when distressed sellers meet lower borrowing costs, so we are keeping dry powder for that window.

The other side of the cycle is constructive. Busts like this tend to leave the infrastructure behind, as the railroad and fiber busts did. Cheap compute and lower rates set up the next phase: AI spreading into small and mid-sized businesses, much as electricity and computing did. That diffusion, not hyperscaler spending, is where economy-wide productivity gains will come from, and it is the realistic path to growing out of the debt. Today's restrictive rates choke that channel; a downturn would reopen it.

Summary

Two engines are carrying the U.S. economy, AI capex and top-decile spending, and both depend on confidence in AI. If financing for the leveraged layer of the AI buildout closes, both stall together. Rates fall, the debt math gets harder, financial repression and higher taxes follow, and the opportunity in real assets comes on the other side.

If you are thinking through how this affects your own portfolio or properties, reach out to us at paladinequitycapital.com.

Data notes

• Revenue comparisons. Anthropic reports gross annualized revenue including cloud-partner sales; OpenAI reports net. Annualized figures extrapolate a recent period and can overstate durable revenue.
• Capex figures. Guidance totals mix company definitions (finance leases, "capital investments") and are directional, not precise.
• Spending share. Moody's (~49%) and BLS (~23%) estimates of the top-10% spending share differ widely; the direction matters more than the level.
• Rates history. The 1956–2025 analysis uses calendar-year nominal GDP growth (U.S. Bureau of Economic Analysis via FRED, series GDPA) and the annual average 10-year Treasury yield (Federal Reserve via FRED, series GS10). The 2026 reading is partial and not comparable. The 10-year is not the Treasury's actual cost of debt, which lags well behind.
• Deficit estimates. The rate-cut and recession figures are rough estimates, not a CBO-style model. The biggest swing factors are the size of any stimulus, how far the long end falls, and whether Treasury shifts issuance toward cheaper bills, which raises savings but also rollover risk.
• Scenarios. Scenarios are judgment calls, not predictions or model outputs.
• Disclosure. Research support for this outlook came from Claude, which is made by Anthropic, one of the companies discussed. Figures comparing AI companies should be checked against primary sources.

Sources

• AOL / Moneywise, top 10% of earners and consumer spending
• BigGo / Financial Times, AI lab revenue accounting and SoftBank financing
• Congressional Budget Office, budget and deficit projections
• Crain Currency, Moody's top-decile spending share
• CreditSights, hyperscaler capex estimates for 2026
• Fortune, OpenAI compute commitments and cash burn
• Fox Business, Moody's on the wealth effect
• Franklin Templeton, BLS view of the top-decile spending share
• ING, AI's contribution to U.S. growth
• Joint Economic Committee, monthly debt update
• Let's Data Science, tokenmaxxing and AI adoption
• Minneapolis Fed, the K-shaped consumer
• Notebookcheck, AI-built Adobe alternatives
• OfficeChai, OpenRouter spend share
• PrimeRates, Treasury debt and average interest rates
• Quartz, OpenAI revenue correction
• Sacra, OpenAI financial projections
• The Wealth Advisor, federal net interest costs
• TweakTown, AI-built Adobe alternatives
• ValueAdd VC, AI capex tracker
• Yahoo Finance, OpenAI 2025 results

Disclaimer

This report is for informational and educational purposes only and is not investment, tax or legal advice, or an offer to sell or a solicitation of an offer to buy any security. Views are the author's as of October 2026 and may change. Scenarios are judgment calls, not predictions. Consult your own advisors before making investment or tax decisions.

Paladin Equity. Protecting What's Real. Pioneering What's Next.