Monthly Newsletter #4: Purity Wall and Line You Never Wrote
Welcome to this month's edition of what I've been thinking about in the technology space. Rather than recap market movements, a few ideas that may seem underweighted in the current conversation.
The Purity Wall
A 2nm chip requires water purer than anything found in nature. We're talking impurities measured in parts per trillion, particles smaller than a virus, water so clean it becomes corrosive because there's nothing left in it.
Here's the problem: as transistors shrink, purity requirements rise logarithmically. Removing 90% of contaminants is standard. Removing the next 90% requires doubling your entire filtration system. Meanwhile, water volume scales linearly with process complexity.
This creates a strange bottleneck. We're approaching the limits of metrology itself: it's actually difficult to measure whether water is pure enough for 2nm. If you can't measure it, you can't certify the cleanroom. So foundries over-engineer redundant systems that take years to calibrate, adding cost and slowing down capacity ramps.
A mega-fab can consume as much water as a city of a million people. Everyone talks about chips and semi fab equipment. The infrastructure underneath the cleanroom remains largely invisible.
The Line You Never Wrote
There's an old engineering saying: the line of code that's fastest to write, that never breaks, that doesn't need maintenance, is the line you never had to write. Software is about managing complexity, not adding to it. Yet the world seems to lack patience for that kind of restraint.
In this new AI era, token metrics are surging. What used to be a flat fee is now being replaced with token consumption. With large companies in an arms race and employees eager to prove their relevance, much of the growth is driven by fear of lagging behind rather than generating true insights. Companies like Uber and T-Mobile are already putting caps on AI spending after seeing costs far exceeding measurable returns. The infrastructure layer is earning outsized returns while downstream applications struggle to justify their capital deployed.
The real promise of AI may not be brute-force productivity or headcount replacement. It may be helping us answer the questions we dare not ask alone: the trajectory of evolution, the boundary between living and not living. That's closer to the original vision of the scientists who started this field than most of what we see today.
If any of this resonates, I'd love to hear from you, and always happy to share more of what we're observing at Shepherd, whether that's infrastructure dynamics or overlooked corners of the stack.
Until next month,
Jess Xu