Categories: Digitalisation

Agentic AI, Robotics, and MANBRIC: The Capex Supercycle Meets the Physical Economy

The AI investment cycle has shifted from software to heavy industry: approximately $600 billion in hyperscaler capex in 2026, increasingly financed through debt, with an emerging risk of circular financing. At the same time, humanoid robotics is industrializing AI in the physical world with overwhelming Chinese manufacturing dominance. The risk is no longer AI disappoints but AI financing goes off the rails.


Key Data
• AI capex by the 4 hyperscalers (Amazon, Alphabet, Microsoft, Oracle): approximately $600 billion in 2026, compared with $309 billion in 2025 (Yardeni) Goldman Sachs estimates $765 billion in 2026 on a broader basis, $1,010 billion in 2027, and $7,600 billion cumulatively over 2026/2031 (reaching up to $1,640 billion per year in 2031).
• Financing shift: debt accounted for 26% of capex in 2025 and approximately 33% in 2026 (GS) Alphabet recorded its first quarter of negative FCF (Q2 2026) Amazon’s FCF fell to $7.6 billion on a trailing 12 month basis S&P downgraded Oracle to BBB (July 2026, due to its OpenAI exposure). Cumulative RPO exceeds $2,300 billion with customer concentration risk (OpenAI, Anthropic).
• Agentic AI in finance: 44% of finance teams will use it in 2026 (vs. less than 7% in Jan. 2025, Wolters Kluwer) approximately $50 billion in global spending on financial services by the end of 2026 (KPMG) 52% of financial institutions are piloting/deploying agentic systems (Cambridge CCAF, 628 organizations) only 14% describe AI as redefining their competitive advantage; 54 AI agent incidents per company in 2025 (IBM).
• Humanoid robotics: 18,500 units delivered in H1 2026 (+272% year over year); China accounts for 97% of deliveries (AgiBot 44%, Unitree 31%); 2026 projections: approximately 60,000 units; 500,000 per year by 2030 (SAG). US players (Tesla, Figure, Agility) remain in the pilot phase (BMW Spartanburg, Amazon).

Market Structure
The cycle is based on a loop: hyperscalers finance capacity leased to AI labs (OpenAI/Anthropic) which commit to multi year RPOs which justify further capex. This is the circular financing that the BIS has publicly likened to past speculative manias. Robotics, by contrast, follows the opposite trajectory: real volumes, falling costs, and Chinese dominance of the value chain (actuators, batteries, assembly) replicating the electric vehicle/solar panel model.


Allocation Implications
• Distinguish between capex beneficiaries (energy, semiconductors, cooling, networks) and capex payers (hyperscalers with deteriorating FCF) the latter face growing credit risk (Oracle BBB as the canary in the coal mine).
• Monitor the hyperscalers’ debt to capex ratio: above approximately 40%, the AI infrastructure debt asset class (and its tokenized version, see Study 1) becomes the next financing channel along with its concentration risk.
• Robotics: the most profitable exposure is probably in the Chinese component supply chain rather than in Western integrators lacking scale.
• Biotech/MANBRIC: generative AI is compressing discovery cycles for molecules and materials (drug discovery, battery materials) an investment theme best accessed through platforms providing tools for R&D (picks and shovels), with caution around valuations unsupported by revenue.
• Asset management: agentic AI reduces banks’ operating costs (approximately 20%, according to McKinsey) but threatens up to $170 billion in profit pools by 2030 if business models fail to adapt favor institutions deploying it with proper governance (the 81% adoption/14% transformation gap is the key screening criterion).


Risks
Correction in the circular financing loop (an AI lab defaulting on its RPO commitments); electricity constraints and permitting delays (capacity pushed back to 2027+) regulatory confrontation between the EU (high risk AI Act requirements) and deployment speed geopolitical dependence on Chinese robotics.

Monitoring Indicators
Quarterly hyperscaler FCF Oracle/CoreWeave credit spreads debt to capex ratio quarterly humanoid deliveries (SAG) actuator prices reported AI agent incident rates RPO and revenue conversion rates.

Oleg Turceac

Recent Posts

RWA & TOKENISATION: THE MIGRATION FROM SPECULATIVE TO INCOME INSTRUMENT

ThesisOver the course of 36 months, the RWA segment has evolved from a speculative NFT…

2 hours ago

Commodities, carbone et smart cities : les actifs réels de la transition programmable

La réalité physique du paradigme numérique : l'IA consomme de l'électricité, les CBDC et la…

2 hours ago

Anthropic: an operational “cash machine”, an accounting “Volatility Machine”

Anthropic's $80 billion in october 2026 A three-tier monetization architecture, dominated by a consumption-based billing…

3 hours ago

Global AI Regulatory Impact & Capital Markets Integration: EU AI Act Timeline & Digital Ratings

As of 16 November 2026, no one has yet been sanctioned under the AI Act,…

6 hours ago

Risk-Off Market Regime and Trading Strategies

Our report describes a pronounced risk-off, higher-for-longer macroeconomic regime on 21 November 2026, driven by…

7 hours ago

Monetary and Regulatory Architecture 2026–2030

The monetary and regulatory architecture of 2026–2030 makes compliance a measurable asset, as four institutional…

1 day ago