Article by Jordi Pumarola Batlle
725 billion dollars.
That’s how much Amazon, Google, Microsoft, and Meta are expected to invest collectively in AI data center infrastructure in 2026.
For context: that figure is roughly equivalent to Argentina’s entire annual GDP. It also represents close to 100% of the combined operating cash flow generated by these four companies — forcing them, for the first time in decades, to tap debt markets to finance growth.
Q1 earnings confirmed three major realities:
→ High-bandwidth memory (HBM) remains in a structural bottleneck. Samsung, SK Hynix, and Micron — which control roughly 90% of the market — have indicated supply constraints are likely to persist through at least 2027. SK Hynix is essentially sold out for 2026.
→ Lead times for power semiconductors have stretched to 30 weeks. The value chain extends far beyond NVIDIA: cooling systems, power infrastructure, advanced packaging.
→ Apple posted its strongest quarterly growth in four years. This cycle is not just a B2B story.
But here’s the nuance that is often poorly communicated:
A significant portion of this thesis is already priced in. Samsung, Micron, and TSMC are trading at price-to-earnings multiples approaching 40x.
Aggregate sector capex already represents 2.2% of U.S. GDP.
This is not the moment to “buy AI” as a broad theme. It’s the moment to calibrate exposure with discipline:
· What is your actual exposure to this theme once you consolidate all the global funds and indexes you already own?
· Are you unintentionally doubling up on the same positions?
· Do you have enough allocation to less correlated segments?
These are the conversations I’ve been having over the past few weeks with clients and professionals looking to incorporate the AI thesis without giving up diversification.
If you work at a multinational company and actively manage your personal investment portfolio, I’d be interested to hear how you’re approaching this cycle.