Chip Stocks Hit a Wall, CXMT Eyes a DRAM Coup, and the Grid Can't Keep Up With AI
We've all heard the hype about AI data centers sucking up insane amounts of electricity. The usual take is "America needs more power plants." But a piece in Works in Progress flipped that narrative on its head — turns out, the U.S. actually has enough electricity. The bottleneck is getting it to where it's needed.
Grid interconnection queues are backed up for years. You can have all the GPUs in the world, but if the transformer station 20 miles away can't handle the load, you're not flipping the switch anytime soon. This feels like one of those boring infrastructure problems that doesn't make headlines but will quietly determine who wins the AI race. Hyperscalers are already sitting on waiting lists just to get grid access.
To be fair, this isn't a new problem — but the scale of AI demand is making it impossible to ignore. A lot of people are wondering if the next big constraint on AI won't be compute or data, but simply a substation permit.
CXMT's Vertical Play — China's DRAM Ambitions Get Serious
ChangXin Memory Technologies (CXMT) is doing something that sounds almost old-school in today's fabless world: they're building a fully vertically integrated supply chain. From raw materials to packaging, all in Hefei. The goal is to nearly triple production capacity and grab 20% of the global DRAM market by 2030.
Now, I've seen Chinese semiconductor companies make big promises before. But CXMT is different — they've actually been shipping DRAM at scale for a couple years, and their LPDDR and DDR5 products are finding real customers. The question is whether they can sustain the quality ramp while scaling this aggressively. SK Hynix and Samsung aren't exactly standing still, and the patent minefield alone would give most startups nightmares.
Still, the market is taking this seriously. If CXMT pulls it off, we're looking at a genuine three-way DRAM price war (four-way if you count Micron). And honestly? More competition in memory is something every system builder should root for.
Wall Street is Quietly Rotating Out of Chip Stocks
The AI trade has been all about hardware for the last 18 months — NVIDIA, AMD, Broadcom, the usual suspects. But cracks are starting to show. Micron dropped 18% in a single week despite posting record earnings. And SK Hynix's rumored $28 billion IPO is creating a very real problem: if institutional investors need to free up cash to buy into the Hynix listing, they're going to sell what they already hold. Guess what they hold a lot of? NVIDIA.
This rotation is subtle but real. Software stocks are regaining ground. The narrative is shifting from "who builds the AI infrastructure" to "who actually makes money from AI." That's a healthier conversation for the industry long-term, but in the short run, it's going to be choppy.
Quick add-on note — the Alphabet vs XLK ETF comparison is worth a look. XLK returned 29% YTD vs GOOGL's 14%, which is a pretty brutal reminder that even when a company crushes earnings (Alphabet's cloud revenue is on fire), stock picking is hard. Sometimes the boring ETF just wins.
Kandou AI Plants a Flag in India
Smaller news, but worth noting: Kandou AI is opening a chip design headquarters in Hyderabad. India's semiconductor ecosystem has been mostly design services and verification work, but we're seeing more serious chip architecture talent flowing there. Hyderabad specifically has been building this reputation for years — the talent pool is real.
- Apple's lobbying Washington to buy cheaper memory from CXMT — a sign that the DRAM shortage is pinching even the deepest pockets
- Autonomous vehicle edge compute market projected to grow from $3.35B to $9.14B by 2030 — if you believe the timeline, which I don't fully, but the trend direction is right
- Vale (the mining company) positioned as an unexpected AI beneficiary — every data center needs copper, steel, and raw materials. Commodity plays are the dark horse of this cycle
The tech landscape this week feels like a transition point. Hardware dominance is being questioned, the grid is the real bottleneck nobody prepared for, and China's DRAM play is getting dangerously real. Not a bad time to be watching from the sidelines with popcorn.
That said — I use Decision Calculator when I need to weigh trade-offs like "should I upgrade my rig now or wait for DRAM prices to drop." It's surprisingly useful for those tech-vs-timing decisions. Nothing fancy, just a clear-headed way to think through choices.
Enjoy the week, everyone.