“The cost of intelligence keeps falling; the hurdle rate for turning it into advantage keeps rising.”
The Great Split is moving from earnings to the cost of capital. AI still has extraordinary growth. What it no longer has is a free discount rate.
What if the next bottleneck of the AI Supercycle is neither chips nor electricity—but the risk-free rate itself? That is the question markets are confronting this morning.
The U.S. 30-year Treasury yield reached 5.3371% on Tuesday, its highest in nearly two decades, before easing toward 5.28% in Asian trading. German 10- and 30-year yields have reached their highest levels since 2011, France's 30-year yield has risen almost 50 basis points since the end of June, and Japan's 10-year yield is approaching 3% for the first time since the mid-1990s. At the same time, Brent is above $91, the Strait of Hormuz remains deeply impaired, and hyperscalers are issuing record amounts of debt to finance AI infrastructure.
The message is not that the AI Supercycle is ending. The price investors are willing to pay for future intelligence is becoming increasingly dependent on the price of present capital. That is a much more consequential regime.
Five Signals Shaping the Next Economy
verified fact and FDK interpretation
-
Long-term borrowing costs are rising across the United States, Europe and Japan even as weaker U.S. employment and consumption have reduced expectations of another near-term Fed move. Investors cite sovereign borrowing, inflation risk from Middle East energy disruption, weaker foreign appetite for Treasuries and heavy corporate financing for AI infrastructure among the forces lifting yields.
The transmission mechanism is changing. The first phase ran AI growth → higher earnings expectations → higher valuations. The next phase adds another chain: AI CAPEX → more debt → competition for capital → higher term premium → higher WACC → lower present value of future profits. The most important AI multiple may therefore become return on invested AI capital ÷ cost of capital — not revenue growth alone. signal: The discount rate can now overpower even exceptional structural growth.
-
Anthropic's annualized revenue exceeded $65 billion at the end of July, up from $47 billion in May and roughly $9 billion at end-2025. Yet the company is simultaneously assembling a revolving bank facility potentially exceeding $10 billion, with banks seeking large commitments partly to position for a future IPO it has confidentially filed for. Alphabet recently raised $25 billion in dollar bonds, roughly $85 billion of equity earlier this year, and is preparing its first Australian-dollar bond.
AI has moved through three financing eras—venture capital, hyperscaler cash flow, global credit markets—and is now entering a fourth: AI-specific financial architecture. That creates a new competitive advantage: financial engineering capable of turning future AI demand into present productive capacity. Banks are becoming part of the technology stack. signal: Financing capacity is becoming a core AI competitive weapon.
-
Tuesday's 5% fall in the Philadelphia Semiconductor Index was violent but analytically useful; European technology also fell 2.5%, with Infineon down 7.6% and Aixtron 8.8%. There was no comparable collapse in underlying demand—Cerebras this morning launched its CS-4 inference system and plans to deploy 600 megawatts of capacity by end-2027.
The market is beginning to separate two concepts it treated as synonymous: AI demand and AI equity returns. Demand can keep rising while multiples fall—whenever earnings velocity lags the increase in discount rate plus expectations. The next phase rewards pricing power, utilization, free cash flow, balance-sheet strength and capital efficiency. This is not the end of the AI trade; it is the end of the easiest version of it. signal: Demand can rise while equity multiples fall.
-
Unitree finished its Shanghai morning session at 883.87 yuan versus a 150.8-yuan IPO price, valuing it around $50 billion; it raised roughly $900 million and was more than 8,000 times oversubscribed by retail. The company is profitable, but few of its robots yet operate in commercial industrial environments. At least six other Chinese humanoid-robotics companies are preparing listings, amid the World Robot Conference in Beijing.
The AI Supercycle is leaving the cloud and meeting the physical world. But embodied intelligence valuation is moving faster than embodied intelligence deployment. Winners will be determined by industrial economics: cost per productive hour × autonomy × uptime × safety × energy efficiency × labor substitution × asset utilization. signal: Physical AI valuation is outrunning physical AI deployment.
-
The STOXX 600 fell 0.69% Tuesday to a two-week low as Germany's 10-year Bund yield reached 3.261%, its highest since 2011. Yet Q2 earnings growth is estimated around 23.4%, and about 12.3% even excluding energy. Europe remains exposed to Hormuz: EU gas inventories were only 57% full earlier this month, a record low for the period.
Europe now contains both sides of the Next Economy—the accelerating (banks, energy grids, defense, industrial automation, rail, Industrial AI) and the decelerating (energy-intensive industry, long-duration tech without cash flow, leveraged and consumption-exposed names). Capital scarcity imposes strategic discipline. Europe's problem is not simply that it needs more investment; it needs investment with returns above a rapidly rising European hurdle rate. signal: Higher Bund yields force a hard sort between Europe's two halves.
Hidden Patterns Beyond the Headlines
recursive dynamics
AI is crowding into the sovereign capital market
Governments and hyperscalers increasingly want the same thing at the same time: long-duration capital. That produces a recursive chain—AI investment → debt issuance → capital competition → higher yields → higher WACC → weaker non-AI investment → greater productivity divergence. The Supercycle can accelerate itself while making capital more expensive for the rest of the economy.
The Great Split is migrating into balance sheets
The first Split rewarded AI equities, the second AI-driven profits, the third may reward superior financing capacity. Capital access determines investment; investment determines productivity; productivity determines cash flow; cash flow determines future capital access. The Great Split is becoming recursive.
Physical AI could be the next Great Value Migration
Unitree's debut reveals enormous appetite for the next application layer: LLMs → inference → agents → Industrial AI → robotics → autonomous physical systems. The scarce assets change—motors, sensors, batteries, precision components, machine vision, manufacturing scale, electricity and field data—potentially broadening the Supercycle far beyond today's semiconductor leadership.
Morning Velocity Reading
FDK analytical readings; not traded market indices
What to Watch Before the Close
five real-time tests
the regime shift in one screen
- AI's next bottleneck may be the risk-free rate itself, not chips or electricity.
- The decisive multiple is becoming return on invested AI capital ÷ cost of capital.
- AI is building its own credit architecture—banks are now part of the technology stack.
- Demand can keep rising while equity multiples fall; the easiest version of the AI trade is over.
- The Great Split is turning recursive, migrating from equity markets into balance sheets.
The question is no longer whether AI creates value—it does. It is who can capture that value after paying the full physical and financial cost of industrializing intelligence. That is why the bond market, not the chatbot, may now be telling us the most important story about the next phase of the AI Supercycle.