Calcus Cal

$MU $TSM $SHMD $ASML The capital intensity required to manufacture High-Bandwidth Memory (HBM) and next-generation interconnect architectures has exceeded the limits of traditional semiconductor capex cycle. CSPs can no longer rely on open-market merchant silicon. They’re actively transitioning into shadow project financiers. By directly funding the physical expansion of memory fabs, hyperscalers de-risk the capex cycle for IDMs but permanently tether the foundry supply chain to their proprietary data center architectures, fundamentally altering the semiconductor capacity planning industry. If you track the supply chain, you’ll see that the big hyperscalers are offering to directly bankroll SK Hynix capacity expansions, converging with TSMC formalizing its 2026 capex guidance at the top-end $56 billion range to meet AI interconnect demands. I think WFE giants like $ASML and $AMAT receive a revenue floor as fab equipment is effectively underwritten by trillion-dollar tech balance sheets rather than cyclical memory pricing, which makes them very interesting to me personally. $SHMD is one my favorite exposure to this. Memory IDMs receive massive off-balance-sheet capital relief, shifting execution risk away from their equity holders. Conversely, hyperscalers must absorb severe capital intensity and depreciation on their own balance sheets, guaranteeing structurally higher COGS for future compute instances.

Great structure on $AMRC, imo. I wanted a pullback and got one today :)

$BLDP $PLUG $BE $EQIX The U.S. power grid infrastructure is aging and has been exhausted, forcing hyperscalers into behind-the-meter power generation. However, deploying massive gas turbines introduces secondary bottlenecks: severe EPA air quality permitting delays and community opposition. Solid oxide fuel cells bypass both the electrical grid interconnection queue and the most stringent emissions permitting. Furthermore, infrastructure private equity is now decoupling the financing of the Bring Your Own Power (BYOP) asset from the data center itself, creating a standalone project finance market for off-grid power generation. Disclosures from the IMN Data Centers Private Equity conference confirm that grid interconnections are structurally delayed up to four years, coinciding with Oracle officially dropping the gas-turbine plan for its massive Project Jupiter in favor of Bloom Energy fuel cells. This imo creates a structural demand for fuel cell OEMs that capture non-cancellable BYOP mandates from hyperscalers desperate to energize gigawatt-scale sites. Midstream natural gas pipeline operators would also benefit here as they'll be needed to supply a steady baseload fuel to these large off-grid fuel cell deployments. Going forward I think hyperscalers face permanent operating expense inflation to fuel these microgrids, but they effectively buy speed-to-market. Securing power without a four-year grid delay allows them to recognize software/compute revenue quicker, expanding their overall margins despite paying a premium for power.

SK HYNIX INTRODUCES INTEGRATED COOLING FOR NEXT-GEN HBM SK Hynix launched iHBM, a thermal management solution for future high-bandwidth memory chips. Instead of relying only on indirect cooling through the core die, iHBM embeds cooling elements inside the HBM package where heat concentration is highest. The company says the technology reduces thermal resistance by 30%. It is designed for next-generation HBM products, including HBM5. SK Hynix says iHBM can operate stably under high-temperature and high-pressure conditions. The technology is also compatible with existing System-in-Package architectures, meaning customers can adopt it with limited design changes.