Opinion · July 2026
Build vs. Buy: What Meta's Chip Plan and Micron's $250B Bet Have in Common
Two headlines landed within 24 hours of each other this week — Micron's $250 billion U.S. manufacturing commitment and Meta's confirmed custom-chip start date — and the market treated them as separate stories, which they are, on the surface. Underneath, I think they're the same bet, made by two different kinds of company, on two different ends of the same supply chain: that the current pace of AI infrastructure spending is durable enough to justify owning the means of production rather than renting them.
The same logic, running in both directions
Micron is a component maker deciding it's worth spending a quarter-trillion dollars to control more of its own manufacturing capacity, rather than staying purely reactive to demand cycles it doesn't control. Meta is a hyperscaler deciding it's worth designing and manufacturing its own silicon, rather than staying entirely dependent on Nvidia and merchant memory suppliers — Micron included. Read one way, these are complementary: more Micron capacity is good for everyone building AI infrastructure, Meta among them. Read another way, Meta's custom-silicon push is explicitly an attempt to need less of what Micron (and Nvidia) sell over time. I own both sides of that tension.
This is the same "build vs. buy" decision every capital-intensive industry eventually faces once a supplier relationship gets expensive or strategically important enough. Airlines eventually decide whether to lease or own their fleets. Retailers decide whether to run their own logistics or outsource to a carrier. What's unusual here is the scale and the speed — $250 billion and 14 gigawatts are commitments normally associated with a decade of planning, both announced inside the same news cycle, both treated by the market as unambiguously bullish for the announcing company.
What I'm actually watching for
Both commitments are multi-year bets that AI-infrastructure demand keeps growing at something close to the current pace. If that assumption is wrong, Micron's $250 billion buildout becomes overcapacity risk in memory pricing, and Meta's custom-silicon push becomes a capital-intensive distraction from its actual advertising business. I don't think either outcome is likely in the near term, but "both companies are betting on the same macro assumption from opposite ends of the supply chain" is a more precise version of the clustering risk already disclosed on the Compliance Ledger than "these are separate stories that happened to land the same week."
One more thing worth naming plainly: this week's genuinely good news for MU, SNDK, and META is a live test of the trading-discipline rules I just added to the Investment Policy Statement — the Blackout Rule and the Holding Period. Good news is exactly the moment sell discipline gets tempting to abandon in favor of chasing a hot headline; the rules exist precisely so a strong week doesn't turn into an impulsive trade.