The purity requirements and chemical sets ('recipes') for manufacturing a chip are first developed by equipment manufacturers, not fabs. Fabs like TSMC then receive and optimize these recipes, creating a dynamic between the toolmaker's guarantee and the fab's continuous improvement process.

Beyond balloons, helium is indispensable for manufacturing semiconductors, launching rockets, and operating MRIs. Its unique properties, like the lowest boiling point of any element, make it irreplaceable in these high-tech applications, including future technologies like quantum computing and nuclear fusion.

For the next few years, the primary constraint on memory production is not a shortage of manufacturing equipment. Rather, it's the physical lack of clean room space. Memory companies, burned by years of low margins, failed to build new fabs, which have a two-year construction lead time.

The seemingly delicate process of chip manufacturing is, at the atomic level, an extremely violent one. This violence requires the most reactive, and therefore most lethal, chemicals for processes like etching and deposition. The danger is a direct specification of the job.

Unlike many AI fields obsessed with compute, the primary bottleneck in materials discovery is the speed and cost of running physical experiments. Progress depends on experimental throughput, not just bigger models or more GPUs.

Leading-edge semiconductor manufacturing requires ultra-pure "six nines" helium. This necessitates a completely separate fleet of specialized liquid containers that can never be contaminated with lower-grade helium. This fractures the already constrained logistics network, creating a fragile "supply chain within a supply chain" for the most critical end-users.

In polymerization processes like DNA synthesis, not all impurities are equal. Bifunctional impurities, which can react at two points, are especially harmful because their disruptive effect is multiplicative as they get incorporated into the polymer chain. This means even trace amounts below 0.1% can ruin an entire batch.

Semiconductor fabs are prevented from stockpiling many critical, hazardous gases by strict on-site storage permits. This creates a reliance on a just-in-time, hyper-reliable supply chain, making any disruption an immediate and existential threat to production.

The manufacturing requirements for AI compute are staggering. Producing the advanced logic and memory wafers for just one gigawatt of data center capacity requires the output of approximately three and a half EUV lithography machines from ASML, representing over $1.2 billion in capital equipment.