AI chips make up 0.2% of all chips sold. This year, they will generate half of all chip revenue.
That is the most lopsided value concentration in modern manufacturing.
Fewer than 20 million chips, out of more than a trillion produced globally, are expected to account for nearly $500 billion in revenue. The semiconductor industry is about to cross $1 trillion in annual sales for the first time. AI is the reason.
The average chip sold last year cost 74 cents. The chips training AI models cost tens of thousands of dollars each. Same industry. Same supply chains. Completely different economics.
This is the thread that connects the headlines. It is why a Google engineer was convicted of espionage for stealing training infrastructure designs, not AI models. It is why Alphabet is spending $175 billion on infrastructure this year. It is why a single plugin release wiped close to $1 trillion from software stocks in days.
The value in AI isn’t in the software. It is in the silicon underneath. And that silicon is scarce.
Memory shortages are projected to drive 50% price spikes by the middle of the year. Foundries running advanced processes are at full capacity. The companies that control this bottleneck (and there are very few) sit at the centre of the most concentrated value chain in technology.
