Vaughn STV

Went through a Starbucks drive-thru in Brantford. Two coffees. I was handed the machine and the first prompt asked if I wanted to tip $1, $3, or $5. You want an 80% tip for handing me coffee??

Musk just launched TERAFAB, which will become one of the largest chip manufacturing plants in the world. Does it require tons of silver? Let’s ask Grok. Yes, a project on the scale of Tesla's Terafab would require a massive amount of silver** — potentially hundreds to thousands of metric tons annually once at full capacity — due to the sheer volume of high-performance AI chips it aims to produce and the silver-intensive nature of advanced semiconductor packaging, interconnects, and system-level hardware. ### Why So Much Silver? Modern AI chips (like those for Dojo, Optimus robots, FSD inference, or custom Tesla/xAI accelerators) rely heavily on silver in non-core areas: - **Packaging and substrates** — Silver pastes, sintering materials, thermal interfaces, and high-reliability bonding (e.g., wire bonding or die-attach) handle extreme power/heat densities. AI chips use 15–20% more silver than traditional processors due to higher currents, denser interconnects, and better thermal/electrical performance. - **Per-chip estimates** — Flagship AI GPUs/processors incorporate roughly **3–11 grams** of silver (typically 5–8g average for high-end designs like H100 equivalents, including substrate, packaging, and related components). - **System/board level** — Full servers/nodes add far more (hundreds of grams per unit) via connectors, power delivery, cooling, and shielding. Terafab's targets amplify this: - **Chip output** — 100–200 billion AI chips per year (Musk's stated goal for the full-scale complex). - **Rough math**: - At a conservative 5 grams per chip → **500–1,000 metric tons** of silver annually just for the chips themselves. - Add 20–50%+ for packaging variability, advanced features, and full ecosystem (e.g., Dojo racks, robot fleets, satellite compute) → easily **1,000+ tons/year** at peak. - **Comparison context**: - Global semiconductor packaging consumed ~1,200–1,500 tonnes of silver in 2024–2025 (18–22% of electronics demand). - AI/data-center growth added 35–50 million ounces (~1,000–1,500 tonnes) of new annual demand recently. - Terafab alone could represent a significant fraction (potentially 10–50%+) of current global semiconductor silver use if it hits scale, dwarfing many existing fabs. ### Broader Implications - **Industrial demand surge** — Silver's total industrial use is already ~650–700 million ounces (~20,000–22,000 tonnes) annually, with AI/electronics as a fast-growing slice. Forecasts show continued growth through 2030 despite some thrifting/substitution in other areas (e.g., solar PV). - **Supply pressure** — Global mine production hovers ~800–850 million ounces/year; deficits persist (5th–6th consecutive year projected for 2026). A single mega-project like Terafab could exacerbate shortages, drive prices higher, or force Tesla to secure long-term contracts/refining partnerships. - **Not just the fab** — The end products (millions of Optimus robots needing chips, Dojo exaflop-scale training, space-based compute) multiply demand across servers, edge devices, and infrastructure. In short: Terafab isn't just building chips—it's building an entire vertically integrated AI/robotics ecosystem at unprecedented scale. That ecosystem is inherently silver-hungry for reliability and performance at the bleeding edge. While Tesla might optimize (e.g., via design tweaks or alternatives where possible), silver remains hard to replace in these high-stress applications. This could make silver one of the key material bottlenecks (alongside wafers, power, talent) for Musk's vision. If Terafab ramps successfully, expect it to become a notable driver of global silver demand.

Happy Friday fellow silver bugs, this is a Bloomberg headline, not a new article I just wrote on the topic Narrator: article eerily similar to the script of last weekly podcast