Not all inflation is caused by an expanding money supply. Imagine waking up tomorrow and discovering that the world suddenly has four billion more people. Food prices would explode overnight, not because farmers forgot how to farm, or because food production disappeared due to drought, but because demand would massively exceed supply.
Inflation doesn’t always come from speculation and money supply, sometimes it comes from…
This is the analogy Jordi Visser uses to explain today’s AI‑driven inflation:
The AI boom has created a sudden, massive surge in demand for a scarce resource, i.e., high‑bandwidth memory (HBM). Just as food inflation emerges when population outstrips agricultural capacity, AI inflation emerges when compute demand outstrips chip and memory capacity.
The AI Boom Created a “Population Shock” in Compute Demand
Over the last 18 months, AI adoption has accelerated faster than any previous technology cycle in history:
- Corporations deploying generative AI
- Cloud providers racing to build GPU clusters
- Startups training increasingly large models
- Autonomous agents running 24/7 workloads
This is the equivalent of adding billions of new “digital consumers” overnight. The only issue is AI “consumers” don’t consume food; instead, they consume compute, memory, and energy.
Compute = the raw processing power (GPU/CPU work) required to run or train AI models.
Compute refers to the processing power, mainly from GPUs and specialized AI chips, that performs the mathematical operations behind AI. When millions of AI agents start working at once, they consume enormous amounts of compute and energy. During the buildout phase (which is just beginning), massive quantities of chips are required to complete the buildout. All this increased demand is inflationary.
The result: Demand for AI hardware is rising exponentially, not linearly.
The Real Bottleneck: High‑Bandwidth Memory (HBM)
Most people think the AI shortage is about GPUs. It isn’t anymore.
The new bottleneck is HBM, the ultra‑fast stacked memory required for training and running large models.
Only three companies can produce it at scale:
And even they cannot keep up.
Why High Bandwidth Memory is So Scarce
- Extremely complex manufacturing
- Low yield rates
- Difficult packaging
- Limited global capacity
- Multi‑year lead times to expand production
This is the equivalent of having only three countries capable of producing food for the entire planet, and suddenly, there are billions more mouths to feed.
How the Shortage Creates Inflation
Inflation occurs when too much demand chases too little supply. This is the classic definition of demand‑pull inflation. This really hasn’t been an issue until recently, except perhaps in some very limited sectors. Historically, we’ve been more concerned with Monetary Inflation, where prices rise because the money supply increases faster than the supply of goods and services. In that case, excess liquidity causes too many dollars chasing the same goods, so prices are bid up. But this demand-pull inflation is an entirely new ball game.
In AI, demand‑pull inflation is resulting in:
1. Rising hardware prices
HBM prices are climbing because supply cannot expand fast enough.
2. Rising cloud compute costs
Rising demand for AI allows cloud providers to increase pricing tiers.
3. Rising training costs
Training a frontier model now costs:
- More money
- More time
- More energy
4. Rising enterprise AI costs
Companies adopting AI face higher:
- Infrastructure costs
- Model‑hosting costs
- Inference costs
Micron’s Role: The “Farmland” of the AI Economy
Micron is now one of the most strategically important companies in the world because it controls a scarce input that the entire AI ecosystem depends on. Micron has a pretty formidable “moat” around its business because it operates in one of the most capital‑intensive and technically demanding parts of the semiconductor industry. Building high-bandwidth memory is not like building CPUs or GPUs. It requires specialized fabrication processes, extreme precision, and billions of dollars in equipment that only a handful of companies on Earth can operate. The barrier to entry is enormous. Even if a new competitor wanted to enter the market, they would be Billions of dollars and years behind, and lack Micron’s proprietary stacking technology.
Just as farmland becomes more valuable when population surges, HBM capacity becomes more valuable when compute demand surges.
Micron’s position is strengthened by:
- Structural, not cyclical, demand
- Multi‑year supply constraints
- Pricing power
- Limited competition
- A global AI arms race
Jordi argues the AI selloff is temporary and the underlying supply‑demand imbalance is still extreme and will remain that way for at least a couple more years.
AI Inflation Is Real Inflation
AI inflation isn’t confined to tech companies. It spills into the broader economy:
1. Higher corporate operating costs
AI is becoming a core business expense, and shortages raise those costs.
2. Higher consumer prices
AI‑powered services (search, productivity tools, creative apps) become more expensive.
3. Higher energy demand
Data centers consume enormous power, raising electricity prices regionally.
4. Higher capital expenditure
Cloud providers and enterprises invest billions to secure scarce hardware.
This is a new inflation vector — digital inflation caused by compute scarcity.
But I Thought AI was Supposed to Be Deflationary?
For years, the dominant narrative has been that artificial intelligence would drive prices down. AI was supposed to automate tasks, increase productivity, reduce labor costs, and make goods and services cheaper. In other words, AI was expected to be deflationary.
And long‑term, that is probably true.
But in the short term, AI is actually creating inflationary pressure, and Jordi’s explanation makes this easy to understand.
AI didn’t just increase demand for compute and memory. It created a sudden, massive surge, which is the equivalent of adding billions of new “digital workers” overnight. These AI agents all need to think, store information, recall past steps, and run complex loops. That requires enormous amounts of compute, memory, and energy.
The problem is simple: Demand exploded far faster than supply can expand.
This is classic demand‑push inflation—too much demand chasing too little supply. The world didn’t suddenly get more GPUs, more HBM memory, or more data centers. Yet AI workloads multiplied almost instantly.
So instead of lowering costs, AI is temporarily raising them:
- Chip prices are rising because memory is scarce.
- Cloud compute prices are rising because capacity is limited.
- Training costs are rising because models are larger and more complex.
- Energy demand is rising because data centers are running hotter than ever.
AI will likely become deflationary once supply catches up, i.e., once memory fabs expand, once data centers scale, once compute becomes abundant. But right now, we’re in the opposite phase.
AI is creating inflation, not reducing it, because the infrastructure required to support it is nowhere near ready for the scale of demand.
This is why Jordi’s analogy works so well: If four billion new people appeared tomorrow, food prices would skyrocket, not because farming got worse, but because demand overwhelmed supply. AI is doing the same thing to chips, memory, and compute.
Key Ideas
- AI was expected to be deflationary, but current demand is rising faster than supply.
- Billions of new AI agents are consuming compute, memory, and energy at unprecedented levels.
- Semiconductor supply chains cannot expand quickly enough to meet this surge.
- High bandwidth memory is the primary bottleneck driving cost increases.
- This creates demand‑pull inflation as too much demand chases too little supply.
- Prices rise not because technology is failing, but because infrastructure is overwhelmed.
- AI may become deflationary later, once chip fabs, data centers, and energy capacity scale.
You might also like:
Image created by: Bing AI
