The AI infrastructure race has created the first genuine physical scarcity cycle in decades. Here’s why that makes Bitcoin’s next move inevitable, and why most investors are still looking in the wrong direction.
There is a version of the current moment that most investors are completely missing. They see the AI story as a technology story, a software story, a story about which model scores highest on benchmarks. That framing is costing them.
The AI buildout of 2025 and 2026 is not a technology cycle. It is a physical scarcity cycle, the likes of which the world has not seen since the post-war industrial reconstruction. And embedded inside that scarcity cycle is the most structurally bullish macro setup Bitcoin has ever had — not because of sentiment, not because of halving mechanics, but because of what happens when you run the largest forced capital allocation in history through an economy that cannot keep pace with the demand.
The numbers alone should recalibrate anyone still watching price charts without macro context.
$725 Billion and the Physical Wall
Google, Amazon, Microsoft, and Meta collectively plan to spend $725 billion on capital expenditure in 2026, up 77% from last year’s record $410 billion. That is not a projection built on hope. Meta went from $13 billion of capex in Q1 2025 to $20 billion in Q1 2026. Alphabet went from $17 billion to $36 billion. Microsoft went from $17 billion to $31 billion. Amazon went from roughly $25 billion to $44 billion.
To understand what is happening here, you need to stop thinking of these companies as technology firms and start thinking of them as industrial builders in a race against physics. Sundar Pichai said Google is compute-constrained in the near term. Microsoft’s CFO attributed roughly $25 billion of its 2026 capex figure to rising memory chip and component costs alone. These are not efficiency problems. They are supply problems. They are the problems of a world where demand has materialised faster than atoms can be arranged into factories.
Since the release of ChatGPT in late 2022, spending on AI chips and data centres has grown exponentially. AI companies have consistently said that they want to expand their compute faster than supply chains allow. AI chip manufacturing is becoming a binding constraint on the pace of the AI compute buildout.
This is the key insight: the bottleneck is not capital. Capital is abundant. The bottleneck is physical production capacity, and physical production capacity cannot be conjured by a press release or a funding round. New factories for conventional chips are not expected to come online until 2027 or 2028, confirming analyst predictions that the memory and GPU shortage will persist well beyond 2026.
The world is not short on money. It is short on foundries, optical fibre, cooling systems, power infrastructure, and the chemicals required to manufacture advanced semiconductors. Despite a huge AI infrastructure buildout over the past few years, available compute is scarce. Anthropic added a staggering $6 billion of ARR in the single month of February alone, driven by broad adoption of agentic coding — and if Anthropic had more compute, they would have added more.
That last sentence is worth reading again slowly. The constraint on AI revenue is not demand. It is supply. We are watching a demand curve hit a vertical wall.
The Population Problem Nobody Is Talking About
Here is a framing that makes the scarcity dynamic visceral. Human population grew from one billion to eight billion over roughly 150 years. AI agent populations are scaling on a timeline measured in months.
DRAM supplier inventories fell to two to four weeks by October 2025, down from 13 to 17 weeks in late 2024, per TrendForce data. That inventory drawdown did not happen because someone forgot to order chips. It happened because the volume of AI inference requests, the tokens these agents consume simply to think and act, grew faster than any supply chain model could have anticipated.
By early 2026, the industry is implementing systems where AI can decide, blockchains can verify, and payments can execute automatically, often with stablecoins and tokenised assets as the settlement layer. This represents the start of a self-coordinating model where software agents can perform economic work without constant human intervention.
Every autonomous agent requires compute. Every compute cycle requires chips. Every chip requires a foundry, a chemical supply chain, a cooling system, and a power grid. The agentic economy is not a software abstraction. It is a physical demand event arriving at speed that physical supply chains were never built to absorb.
Worldwide semiconductor revenue is expected to reach $800 billion in 2025, growing 17.6% year-over-year. Datacenter semiconductors remain the primary growth driver. Demand for AI infrastructure and accelerated computing is fuelling significant semiconductor revenue expansion.
And yet this is, by every structural measure, the early innings of inference demand. The training phase of AI, the phase the world just spent three years and hundreds of billions completing, is not where the sustained compute appetite lives. That appetite lives in inference: the constant, continuous, around-the-clock process of agents thinking, deciding, transacting, and iterating. Training happens once per model. Inference happens billions of times per day, and the population of agents placing inference requests is only beginning to grow.
Why This Is Macro, Not Tech
The mistake most portfolio managers are making is categorical. They are filing AI under “technology sector” and applying the analytical toolkit they built for software stocks: TAM models, net retention rates, recurring revenue multiples. That toolkit is wrong.
Capex as a percentage of sales will continue to climb to seemingly untenable levels in 2026, including roughly 86% for Oracle, 54% for Meta, 47% for Microsoft, and 46% for Alphabet. No technology company in history has sustained capex at these ratios for multi-year cycles. What these numbers describe is not a software company investing in growth. They describe an industrial builder investing in physical plant, with all the implications that carries: long lead times, pricing power at the bottleneck, and multi-year return horizons.
The investable themes that emerge from this are not about picking the best AI model. They are about identifying where the physical constraints concentrate. Liquid cooling. Power semiconductors. Specialty chemicals used in chip fabrication. Optical fibre for the data centre interconnects that agentic workloads demand. High-bandwidth memory. These are commodity bottleneck plays in a world where most institutional capital is still searching for the next SaaS multiple.
Jet engine shortages are threatening AI data centre expansion, with wait times stretching into 2030, as the rush to power the AI buildout continues. Jet engines. The same industrial supply chain constraints that limited aerospace capacity are now limiting the power generation systems that data centres require. This is not a software problem. This is a 1970s-style physical capital cycle, compressed into a timeline that would have been unthinkable in any prior era.
The Convergence: Where AI Scarcity Meets Bitcoin’s Monetary Thesis
Now comes the part that most people in both the AI conversation and the crypto conversation are treating as separate stories, when they are, in fact, the same story.
The AI buildout has done something remarkable and largely unappreciated: it has created the conditions for the single most compelling macro trade in Bitcoin’s history, and it did so through a completely non-crypto mechanism.
Follow the logic.
The hyperscalers are spending $700 billion in 2026, with capex now exceeding internal cash generation, forcing hyperscalers to debt markets. Projections from Morgan Stanley and JPMorgan suggest the technology sector may need to issue $1.5 trillion in new debt over the next few years to finance AI infrastructure construction. That debt issuance happens against the backdrop of a US government that is already running structural deficits and a Federal Reserve that is trapped between sticky inflation and slowing growth.
The yield on the 30-year US Treasury note rose to 5%, hitting its highest since July 2025. Rising Treasury yields are sucking capital out of risk assets. The Energy Information Administration described the wider Brent-WTI spread and disrupted Strait of Hormuz navigation as part of the global crude-market backdrop, which keeps the inflation channel open and the Fed defensive into the back half of the year.
Three-month Treasury bills yielding less than inflation. Long bonds cracking 5%. A Fed that cannot cut without reigniting price pressures it never fully extinguished. This is the negative real yield environment that has historically generated Bitcoin’s most dramatic returns. Not because of narrative. Because of mathematics: when the risk-free rate fails to compensate for inflation, capital seeks alternatives, and in a world where traditional alternatives, private equity, private credit, real estate, have underperformed for three consecutive years, the alternatives menu is shorter than it has ever been.
Fiat currencies and assets denominated in fiat currencies face additional risks due to high and rising public sector debt and its potential implications for inflation over time. Scarce commodities, whether physical gold and silver or digital Bitcoin and Ether, can potentially serve as a ballast in portfolios for fiat currency risks. As long as the risk of fiat currency debasement keeps rising, portfolio demand for Bitcoin and Ether will likely continue rising as well.
The AI capital cycle is, paradoxically, one of the most powerful forces accelerating this dynamic. The more capital pours into physical infrastructure at scale, the greater the pressure on government balance sheets, corporate debt markets, and ultimately the purchasing power of the currency in which all of that debt is denominated. The physical scarcity of chips and compute creates monetary expansion pressure that ultimately lands in hard asset prices.
The Supply-Side Story Nobody Is Writing
In April 2026, US spot Bitcoin ETFs absorbed approximately 19,000 BTC over a nine-day streak, nine times the amount of new Bitcoin mined in that same period. This supply crunch is creating mechanical upward pressure on price, independent of retail speculation or hype-driven rallies.
The structural setup here is quietly extraordinary. BlackRock’s IBIT alone captured an estimated $2.1 billion to $3 billion of April’s inflows, growing its holdings to approximately 812,000 BTC, roughly 3.8% of total Bitcoin supply and placing it in the top 1% of all US ETFs by flow metrics.
Americans aged 60 and over hold about $110 trillion in net worth, a figure that will shift to younger generations over the coming decades. This projected wealth transfer represents one of the biggest financial changes in modern history, with industry estimates placing the total between $84 trillion and $124 trillion by 2045 to 2048. Younger generations show a starkly different investment profile, with 45% of Gen Z and Millennials owning crypto compared to just 18% of Gen X and Boomers.
Despite all of this, less than 0.5% of US advised wealth is currently allocated to the crypto asset class. That number, against the backdrop of $110 trillion in intergenerational wealth transfer and the structural macro conditions described above, is the most important asymmetry in global markets right now. The question is not whether this allocation grows. The question is at what pace.
Agents, Wallets, and the Infrastructure Bitcoin Was Built For
There is a second structural bid for Bitcoin and crypto infrastructure that operates entirely outside the traditional investment thesis, and it may ultimately dwarf the institutional allocation story.
Inference costs represent 23% of revenue at AI business-to-business companies. These costs do not decrease as the platform scales. AI agents need to pay for compute. They need to pay for data access. They need to settle transactions with other agents at machine speed, at volumes and frequencies that make conventional financial rails structurally incapable of serving the demand.
Machine-to-machine economic activity operates at fundamentally different scales. An agent might pay $0.001 for a single inference, $0.00001 for accessing a database record, or $0.0000001 for storage of a small data fragment. These transactions occur at frequencies that humans cannot match. Blockchain networks, particularly Layer 2 solutions, can support high-frequency micro-transactions with fees measured in fractions of a cent.
The agentic economy does not need a narrative for crypto. It needs crypto the way a combustion engine needs fuel. The programmable, permissionless, globally accessible financial rails that crypto provides are not a nice-to-have for a world running on autonomous agents. They are the only architecture that can handle machine-velocity settlement without centralised intermediaries creating bottlenecks at every junction.
The velocity at which AI operates demands financial infrastructure that can keep pace. Traditional financial systems, with their hours- or days-long settlement times, are increasingly becoming bottlenecks. Cryptocurrencies, particularly stablecoins like USDT and USDC, along with Bitcoin as a foundational settlement layer, provide the high-speed, borderless rails necessary for AI agents to transact at machine velocity.
This is not a speculative future state. AWS’s AgentCore Payments infrastructure, built on Coinbase and Stripe rails and settling in USDC across Base and Solana, is live today. The agent economy is not coming. It is running.
The Positioning Gap
Here is where the analysis becomes actionable.
The institutions most capable of moving markets are the most structurally impeded from participating. Pension funds are benchmarked against bond yields. Hedge funds face drawdown constraints incompatible with the volatility profiles of assets growing 5 to 10 times in a cycle. Mutual funds move through due diligence processes built for companies that change quarterly, not markets that reprice weekly.
The macro-sensitive sellers, the cohort that historically reacted to every interest rate hike, every CPI print, every geopolitical headline, have already exited the market. What remains is a base of conviction holders, ETF accumulation that is nine times mining supply in monthly flow, and a macro environment that is generating negative real yields with no clear path to resolution.
The same scarcity logic that applies to AI chips applies to Bitcoin. You cannot manufacture more. The demand curve is moving. The supply curve is fixed.
CoinShares projects Bitcoin trading between $110,000 and $140,000 in a base case of subdued growth and sticky inflation. The bull case sees Bitcoin climbing above $150,000 if AI-related productivity gains allow the Fed to cut rates more decisively.
The framework for thinking about this moment is not 2000 or 2017 or any prior cycle. It is the 1970s, reimagined with programmable money, autonomous agents, and a physical capital cycle running at software speed. Scarcity is the trade. Not the scarcity of ideas, but the scarcity of atoms, of foundry capacity, of mined coins, of real purchasing power in a world where governments are financing the most expensive infrastructure buildout in human history with currencies that lose value faster than the assets they are funding gain it.
The window to position in front of that is open. The question is whether you are looking at the right scoreboard.
Trade the crypto markets on exchanges built for derivatives and spot with deep liquidity. Our recommended platforms for serious traders include Bybit, BloFin, and OKX. For decentralised perps and on-chain leverage, gTrade remains one of the most capital-efficient platforms available.
Recommended reading:
Tokenized US Treasuries: Best On-Chain Yield Products Compared in 2026
The CLARITY Act and the $30 Trillion Gate: What Actually Happens to Bitcoin Now
Sound Money Wins Every Century: A 500-Year History of What Holds Value and What Collapses
Real-World Lessons on Currency Collapse, Capital Controls & the Rise of Bitcoin and Stablecoins
Crypto Friction Cost Index 2026: What Active Crypto Traders Actually Pay in Hidden Costs Every Year
The Pandemic Market Playbook: How Outbreaks Move Stocks, Real Estate and Crypto
How to Use Bitcoin as Collateral for a Loan in 2026: The Complete Guide to Crypto-Backed Lending
Crypto for Retirement: The Complete 2026 Guide to Building a Self-Directed Crypto Pension
Bitcoin vs Gold vs Real Estate vs S&P 500: 10-Year Return Comparison
How to buy Bitcoin in 2026: the complete beginner’s guide by country
Lowest Crypto Futures Fees in 2026: Every Major Exchange Ranked and Calculated
Best Crypto Sign-Up Bonuses and New User Rewards in 2026
Best Crypto Futures Trading Competitions in 2026: Where Active Traders Can Earn Rewards
About the Author
