The global memory shortage is something that will probably end, perhaps even soon. e.g. Google announced Turboquant[1] today. If as described, it would significantly reduce thirst for more RAM at data-centres. Even if it doesn't pan out, data centres aren't being built at the same rate as they're being financed due to the practical, but unavoidable, problem of finding enough power. Demand for memory may actually have a very real bottom that we hit soon.
However, there is another reason to look after and hang onto to certain types of products long-term.
Tariffs.
If the trade barriers that the Trump administration has put up remain long-term, it fundamentally changes what can be built. High volume items (like RAM) are the least likely to be affected. Low-volume, high performance items are what are threatened. Say you're building a very specialized, very low demand item that's simply-the-best. You're probably going to source the best components and materials from several countries, build it in one place, and then ship it globally. You amortize the cost of tooling, etc. across the entire global market.
If a few countries throw up trade barriers, as the U.S. has done, your material costs go up and your access to markets decreases. People on the other side of those trade barriers may suddenly not be able to afford your products. Supply gets more expensive and demand drops. What was marginally profitable in the old world order becomes uneconomic in the new order. Such items aren't going to be magically on-shored to the U.S.. They're just not going to be made anymore.
If you own something that's niche and barely profitable to make, that's what you should look after and take care of, because more of it might not be made for a while if trade barriers don't come back down.
So it's been public knowledge for almost year, and internal Google knowledge for longer. Did Google reduce its capital expenditure plans over the past year? Let's ask Google:
Google is expecting to see a capex of between $175-185 billion in 2026, approximately double that of 2025.
If turboquant can reliably reduce LLM inference RAM requirements by 6x, suddenly reducing total RAM needs by 6x should have a dramatic shift on the hardware market, or at least we can all hope. I know 6x is the key-value cache saving, so I'm not sure if that really translates to 6x total RAM requirements decrease for inference.
However, there is another reason to look after and hang onto to certain types of products long-term.
Tariffs.
If the trade barriers that the Trump administration has put up remain long-term, it fundamentally changes what can be built. High volume items (like RAM) are the least likely to be affected. Low-volume, high performance items are what are threatened. Say you're building a very specialized, very low demand item that's simply-the-best. You're probably going to source the best components and materials from several countries, build it in one place, and then ship it globally. You amortize the cost of tooling, etc. across the entire global market.
If a few countries throw up trade barriers, as the U.S. has done, your material costs go up and your access to markets decreases. People on the other side of those trade barriers may suddenly not be able to afford your products. Supply gets more expensive and demand drops. What was marginally profitable in the old world order becomes uneconomic in the new order. Such items aren't going to be magically on-shored to the U.S.. They're just not going to be made anymore.
If you own something that's niche and barely profitable to make, that's what you should look after and take care of, because more of it might not be made for a while if trade barriers don't come back down.
----------
[1]https://research.google/blog/turboquant-redefining-ai-effici...