Key points
- 01The 2026 AI cycle has become a physical investment cycle in data centers, servers, memory, networking, power, and construction—not only a software narrative
- 02Direct beneficiaries require confirmation through orders, pricing, utilization, cash flow, and customer concentration rather than revenue growth alone
- 03Growth in BIS dollar and euro credit is inconsistent with a simple liquidity shortage, while elevated valuations and hedge-fund leverage leave amplification risk
- 04The market still resembles a compressed regime around AI leaders and suppliers; equal-weight performance, earnings breadth, and credit can confirm a transition
- 05A cycle end is better identified through sustained deterioration across orders, exports, estimates, margins, credit, and breadth than through one price decline
Chapter 1. What era are we living through?
An AI arms race
AI competition in 2026 is a capital race for compute. Microsoft said it expected roughly $190 billion of calendar-2026 capital expenditure, including higher component pricing, while Meta guided to $130–145 billion including principal payments on finance leases.
Spending does not prove return. New capacity raises depreciation, power, network, and component cost; free cash flow and margins can weaken even while revenue grows.
From the internet to AI and physical AI
The internet lowered information-distribution cost. Generative AI is reducing the marginal cost of producing language, code, and media. Physical AI connects models to sensors, robots, vehicles, and factories, where latency, safety certification, maintenance, energy, and liability can slow deployment.
Robotics is likely to begin in repetitive, hazardous, standardized environments. Utilization, human interventions, installation cost, service burden, and repeat purchase are more informative than demonstrations.
Programmable payments and tokenized money may reduce settlement and working-capital friction, but only when compliance, fraud, reserves, interoperability, and finality remain robust. Digital-asset prices are not themselves evidence of economy-wide productivity.
Why big technology companies led
Large platforms combine cash flow, distribution, data-center experience, and integrated model-to-product software. Those advantages fund long lead-time investment, but model commoditization, power constraints, regulation, customer multi-cloud strategies, and depreciation can still lower returns.
Amazon reported that AWS AI reached an annual revenue run rate above $25 billion in Q2 2026 while also identifying AI investment as a major reason trailing free cash flow fell. The same spending can create revenue and balance-sheet pressure.
AI capex redraws resource allocation
Capital moves from cloud buyers to chip design, foundries, memory, networking, cooling, power equipment, and construction. TSMC reported $40.2 billion of Q2 2026 revenue, reaching the high end of guidance. Yet bottleneck suppliers can later face overcapacity, and long contracts can exchange visibility for customer concentration.
Chapter 2. Where are this cycle's leaders?
Direct AI-capex beneficiaries
Candidates need improving backlog, pricing, utilization, prepayments, and cash flow—not merely exposure to servers. Power conversion, cooling, optical networking, memory, and semiconductor equipment can be as important as accelerators. Supplier capacity spending means revenue growth must be tested against margin, working capital, and post-expansion asset turnover.
The surge in East Asian exports
Korea's trade ministry reported July 2026 exports of $98.89 billion, up 62.8% year over year, with semiconductor exports up 178.8% to $41.01 billion. The strength is significant, but base effects, price versus volume, inventory building, and tariff front-loading must be separated before extrapolation.
Mid-cap operating leverage
Component and equipment companies can grow profit faster than revenue after utilization clears break-even. The reverse also holds. Fixed cost, customer concentration, inventory, receivables, and expansion debt distinguish structural advantage from a temporary profit spike.
The dot-com lesson and today's differences
The entity funding infrastructure does not always earn the highest shareholder return; scarce components, standards, software, and customer relationships may capture more value. But the analogy is incomplete because today's hyperscalers generally possess stronger cash flow and installed customers than many telecom entrants did.
Why leadership matters and how a cycle ends
Capex travels through orders, production, exports, revenue, earnings, and cash flow. A cycle end is a sequence: slower orders, rising inventory and receivables, estimate cuts, weakening prices, and wider credit spreads. Long-term quality ownership should not become permanent ownership of normalized cyclical earnings.
The structural tests are AI revenue and productivity versus depreciation and power cost; external customer adoption versus internal workloads; bottleneck relief without price collapse; and debt-funded spend that remains serviceable at higher rates.
Chapter 3. How to read the direction of money now
Broad liquidity is a river, not one balance sheet
Broad liquidity combines bank lending, bond issuance, reserve-currency credit, fiscal flows, and collateral. BIS reported dollar credit to non-bank borrowers outside the United States at $14.743 trillion in Q1 2026, up 7.3% year over year. The destination, purpose, and price of credit matter more than the aggregate alone.
Narrow liquidity is market plumbing
Narrow liquidity concerns tradable cash, dealer intermediation, bid depth, collateral, leverage, and redemption structures. The Fed's May 2026 report assessed bank funding risks as moderate but hedge-fund leverage near record highs. Aggregate liquidity can coexist with a fragile exit.
Liquidity is concentrated
The IMF identified stretched valuations and AI-related equity concentration as downside vulnerabilities. This looks less like an absence of money than a concentration of risk appetite. Cap-weight versus equal-weight performance, participation, credit, and smaller-company strength reveal whether it is broadening.
What are Bitcoin and gold waiting for?
Bitcoin responds to dollar liquidity, real rates, risk appetite, regulation, and leverage liquidation. Gold also responds to reserve demand and geopolitical risk. The World Gold Council reported 289 tonnes of Q2 2026 central-bank net purchases alongside 45 tonnes of gold-ETF outflows. Long-horizon reserve demand and short-horizon investor flows can point in opposite directions, so neither asset is a single liquidity gauge.
Chapter 4. The moment money changes direction
The yield curve: the reason for steepening matters
Bull steepening driven by falling short rates during easing differs from bear steepening driven by rising long yields, inflation, fiscal risk, or term premium. Stable credit and earnings can make the former supportive; rising long rates, a stronger dollar, and wider spreads can make the latter restrictive.
Is 47 basis points an absolute threshold?
No threshold has economic meaning without the maturity pair, sampling interval, period, and regime from which it was derived. The speed and cause of curve movement and confirmation from credit, earnings, and inflation matter more than a standalone 47 bp level. Bands and persistence reduce repeated flips around measurement noise.
Geopolitics and rates
A geopolitical shock can lower yields through safe-haven demand or raise long yields through energy inflation, fiscal spending, and supply disruption. Observe inflation expectations, the dollar, oil, Treasury auctions, and credit spreads rather than reasoning from the event label alone.
Market breadth: why the index rises without a portfolio
Cap-weighted indices give the largest firms the greatest influence, so an index can rise while equal weight, smaller firms, and advance counts lag. A lagging holding may be overlooked, but it may also lack estimate revisions, financial strength, or sector flows; the thesis still requires review.
Compression and broadening portfolios
A compression portfolio owns validated leaders and direct bottleneck suppliers under tight stock caps. A broadening portfolio expands toward equal weight, mid-caps, and lagging sectors only after earnings participation improves. More cheap stocks without broader evidence is not necessarily diversification.
When to move from compression to broadening
Evidence strengthens when equal-weight relative performance, the share of stocks above medium- and long-term trends, the breadth of estimate upgrades, and small-company credit conditions improve for several months. Gradual rebalancing reduces the cost of a false breakout.
Current positioning and Korea's two-stage broadening
As of August 29, 2026, the evidence shows powerful AI capital spending and East Asian supply-chain results alongside elevated valuation and concentration risk. A reasonable framework is to retain a capped leader core while increasing peripheral exposure only as breadth confirms, with cash and short-duration instruments preserving optionality.
Korea's first broadening stage runs from semiconductor exports and equipment imports into large exporter earnings. The second requires wages, domestic capex, services, and non-IT mid-cap estimates to improve. Export totals alone cannot establish that second stage.
The direction of money appears in linked balance sheets, not one chart
AI capex is simultaneously a cloud company's cash outflow, a supplier's revenue, a utility's investment requirement, and a bond market's financing need. A benefit on one balance sheet can be a capital burden on another.
Current evidence shows strong demand and expanding credit, not a guaranteed future return. High valuations, concentration, and leverage imply that a small disappointment can create a large price response.
A practical monthly dashboard can place hyperscaler capex guidance, Korean and Taiwanese exports, BIS credit, credit spreads, equal-weight relative strength, and earnings-upgrade breadth on one timeline. The information cutoff for this article is August 29, 2026, and it is not investment advice.
Epilogue: companionship, not perfection
No model is perfect. Markets create industries and rules, companies change accounting and capital allocation, and each investor's horizon and circumstances evolve. One score and target price cannot permanently explain those changes.
A durable edge comes less from never being wrong than from detecting error early while preserving the ability to make the next decision. Timestamp data, separate inference from facts, write falsification before purchase, and evaluate outcomes separately from decision quality.
This series is intended as a companion for updating questions, not a system that decides for the reader. Simplify the model for the available universe, cost, and risk budget, and do not allocate capital to a weight that cannot be explained.
Appendix 1. Core concepts of the Macrobound investment philosophy
Macrobound here means a framework connecting macro conditions with company analysis. It is an independent translation of the questions in the supplied table of contents, not a reproduction of a book's proprietary full definition.
Its seven concepts are: capital moves unevenly along constraints and expected returns; broad and narrow liquidity are separate; market, sector, and company strength are checked in sequence; value, growth, momentum, and quality answer different questions; price is cross-checked with financial resilience; valuation is managed as an assumption range; and a portfolio is designed to survive error and retain optionality.
The causal path is macro liquidity → industry capex → company orders and earnings → price and breadth → portfolio weight. A break in any link prevents a favorable upstream event from being treated as guaranteed investment return downstream.
Appendix 2. Investment glossary
| Term | Meaning | Caution |
|---|---|---|
| Leader | A company where capital, fundamentals, and relative strength converge | Not identical to a hindsight top performer |
| Factor | A shared characteristic explaining differences in return and risk | Results depend on definition and sample |
| Percentile rank | Relative location inside a peer group | Removes the absolute gap |
| COMP | This series' value-growth-momentum-quality comparison model | Not a standard index or return guarantee |
| F-score | Piotroski's sum of nine binary accounting signals | Originally designed for high book-to-market stocks |
| GPM | Gross profit margin | Price, mix, and direct cost require separation |
| DCF | Present value of expected future cash flow | Sensitive to terminal value and discount rate |
| Reverse DCF | Infers the growth and margin required by today's price | A test of assumptions rather than a target |
| Broad liquidity | Credit, bonds, money, and fiscal financial conditions | Different from market-level tradeability |
| Narrow liquidity | Bid depth, dealers, collateral, and redemption plumbing | Normal observations can understate stress |
| Steepening | A widening long-short yield spread | Bull and bear causes have different meanings |
| Breadth | The range of stocks and industries participating | Cannot be inferred from cap-weight index return alone |
| Compression | Performance concentrated in a few large leaders or suppliers | Exit and single-thesis risk increase |
| Broadening | Price and earnings improvement spreading to more groups | Must be separated from a short laggard bounce |
Appendix 3. Rationale for quantitative model weights
Logical and empirical grounds for allocating COMP and F-score
The models are allocated to observe different time horizons. COMP reads changes in current expectations through price, estimates, growth, and quality. Extended F-score tests whether already-reported profitability, cash flow, balance-sheet safety, and efficiency can support those expectations.
Evidence supplies direction, not exact weights. Value, momentum, profitability, and Piotroski signals have shown explanatory power in different samples, but realized results vary with market, period, cost, and data construction. The weights below are precommitted defaults designed to reduce overlap and modestly favor faster information, not coefficients copied from a paper.
Three principles for setting weights
The first principle is independence: do not count the same price or earnings signal several times under different names. The second is stability: small changes in window or threshold should not completely overturn the ranking.
The third is implementability: account for release lags, missing data, turnover, and trading cost, and remove untradeable securities before scoring. A simple equal-weight benchmark takes priority over an inexplicable optimized solution.
COMP weights: the rationale for 40/30/20/10
The COMP default is 40 momentum, 30 quality, 20 growth, and 10 value. Momentum receives the largest weight because it reflects the current direction of information and capital fastest. Quality is second because it reduces exposure to fragile trends.
Growth checks the economic basis for the trend but carries reporting and estimate lags. Value is a smaller guardrail against extreme expectations so it does not mechanically remove every early high-growth leader.
| Area | Weight | Example | Primary failure |
|---|---|---|---|
| Momentum | 40 | Medium-term relative return and estimate revisions | Sharp reversal and turnover |
| Quality | 30 | Cash conversion, profitability, and leverage | Overpaying for a good company |
| Growth | 20 | Revenue, earnings, and cash-flow growth | Base and acquisition effects |
| Value | 10 | Peer multiples and reverse expectations | Value traps and industry mismatch |
Extended F-score weights: the rationale for 40/30/20/10
The original Piotroski score gives one point to each of nine signals. The extension here does not replace it; it uses continuous observations to sort ties with 40 profitability and cash conversion, 30 balance-sheet safety, 20 operating efficiency, and 10 dilution and accounting risk.
Profit and cash test self-funding most directly. Balance-sheet safety tests survival, operating efficiency tests the direction of margin and asset use, and dilution and accounting risk test the reliability and ownership of the reported result.
| Area | Weight | Observation | Relation to the original |
|---|---|---|---|
| Profit and cash | 40 | ROA, operating cash flow, and accruals | Adds magnitude to four profitability signals |
| Balance-sheet safety | 30 | Leverage, liquidity, and coverage | Supports leverage and liquidity signals |
| Operating efficiency | 20 | GPM and asset-turnover change | Adds magnitude to two efficiency signals |
| Dilution and accounting | 10 | Share count, audit, and one-offs | Supports issuance and reliability review |
Top-decile GPM improvement: the quality of the profitability structure
The top 10% of industry-relative GPM improvement identifies companies with unusually strong gross-margin change. Gross margin sits before selling, administration, and financing costs, so it helps reveal changes in pricing, mix, and direct input cost at the start of operating leverage.
Top decile is a research priority, not a buy signal. Falling inputs, one-time price increases, divestitures, and accounting reclassification can all lift GPM, so volume, price, mix, unit cost, and persistence require review.
Weights are not fixed constants
The 40/30/20/10 splits are operating defaults rather than natural laws. Financials, early growth companies, and resource firms may require separate structures; sparse data may favor fewer indicators or equal weights.
Change weights on independent forward evidence, after-cost results, drawdown, candidate stability, and explainability—not recent return chasing. Preserve the date, reason, and prior results so the model can learn rather than rewrite its history.
Sources
Related notes
Direction of MoneyReading the market through an investor's eyesExplains how a few long-run winners create most market wealth and how value, growth, momentum, and quality can frame a search for future leaders
Direction of MoneyThe craft and philosophy of finding market leadersDesigns a reproducible process for reading market, sector, and stock strength and ranking candidates with percentile-based multi-factor scores
Direction of MoneyTesting the financial health of market leadersConnects Piotroski's accounting signals with composite rankings, valuation ranges, and an integrated sell disciplineThis is a personal research note, not investment advice