Semiconductor analyst 'Bubble Boy' says advanced packaging is the new Moore's Law — and the market hasn't priced it in

Advanced packaging is the new Moore's Law

Bubble Boy, a semiconductor analyst who runs a day job in the industry, argues that the market is systematically mispricing advanced packaging — and that this is the most important structural shift in semiconductors right now.

The core argument is that shrinking transistor nodes, the dynamic that defined Moore's Law, is no longer the primary driver of compute scaling. What matters now is how many dies you can stack and interconnect in a single package. In 2025, Nvidia consumed 60% of TSMC's advanced packaging capacity, leaving every other chip designer to compete for the remainder. That constraint is already shaping product roadmaps: Broadcom, for instance, had to abandon a four-die package configuration at the last minute. Bubble Boy sees this dynamic intensifying as hyperscalers push for larger dies, more HBM, and denser integration — all within one package.

“Advanced packaging to me is the new Moore's Law... In 2025, NVIDIA was 60% of TSMC's advanced packaging, so everyone else was out there to fend for themselves. Any kind of advanced packaging that comes to market allows people to scale compute way more than just shrinking a node... I think we have at least three years of CapEx growth going crazy.”

His Intel thesis has shifted accordingly. The bull case used to rest on the fab turnaround, specifically the 18A and 14A node ramp. Now he thinks the real opportunity is Intel's advanced packaging capabilities, which he believes are underappreciated relative to where the industry is heading.

Three more years of CapEx growth is his base case. Every major hyperscaler has telegraphed roughly $680 billion in combined capital spend, and almost none of the data center buildout has actually been completed. He treats the current moment as the beginning of the cycle, not the middle.

Flash memory is his other conviction. He hints at a career move into the flash supply chain — either with a large established company or an earlier-stage player — and argues that flash scales further than the market expects. On near-term earnings, he thinks Sandisk can keep raising flash prices until customers push back, and expects this earnings season to deliver a clear verdict on whether customers will absorb CPU price increases of 25 to 30%.

One caveat worth flagging: Bubble Boy demonstrated, apparently as a point about LLM-generated research risk, that he could prompt a model to frame a random Japanese chemical supplier as an AI bottleneck. He says the company in question is not actually a bottleneck. The episode underscores a real risk in the current market — AI-assisted research can produce convincing-sounding theses that move retail money into positions built on nothing.

His edge, as he frames it, is working on the same engineering problems the companies he covers are trying to solve. Financial analysts, he argues, understand the P&L implications of yield and packaging but not the competitive ones. That gap is where he thinks the alpha lives.