Gold, chips and AI capex

Goldman Sachs has been making two linked bets about the economy: one about real assets and one about the silicon under AI. The bank says gold is no longer just a safe‑haven trade but a structural holding, and it is also forecasting that AI‑driven capital spending will send semiconductor revenues far higher over the next year. (goldmansachs.com 1) (goldmansachs.com 2) The hard‑asset claim is simple. Goldman highlights growing demand from central banks and private investors for gold as a portfolio diversifier and hedge against concentrated tech risk, and it has raised its numerical target for late‑2026. (kitco.com) (goldmansachs.com) The chip call is equally concrete. Goldman’s research group estimates that hyperscale AI capital expenditure could top half a trillion dollars in 2026 and that AI‑related demand will push semiconductor revenues sharply higher—about a 49% jump from current levels by the end of 2026 in the bank’s read of the market. (goldmansachs.com) (economictimes.indiatimes.com) Put those pieces together and a picture emerges: software‑heavy AI models are creating durable demand for physical hardware—GPUs, accelerators, high‑bandwidth memory, servers and the racks and power plants that host them. Hyperscalers (the largest cloud and AI companies) are buying enormous compute fleets, which in turn lengthens the revenue runway for chipmakers and the industrial suppliers that build data‑center gear. Goldman quantifies that capex in public forecasts and investor notes, and then markets can price sectors that feed that demand differently than they price pure software names. (goldmansachs.com) The industrial signal is already visible on the ground in Ohio. Foxconn, which had bought the former Lordstown Motors factory, sold the site and is now converting it into a plant to make AI server and data‑center equipment under a SoftBank‑led Stargate initiative. The facility was reported sold for roughly $375 million, SoftBank has announced Stargate sites including Lordstown, and Foxconn says renovation work and joint manufacturing plans are underway. (bloomberg.com) (datacenterdynamics.com) (businessjournaldaily.com) For a student building econometric and trading projects, these are actionable links between macro forecasts and cross‑sectional asset moves. One project: build a time‑series nowcast of semiconductor revenue using hyperscaler quarterly capex, GPU shipments, and spot GPU price spreads, then test whether that nowcast improves returns on a long‑short semiconductor‑equipment ETF strategy versus a software growth basket. Another: run an event‑study on regional suppliers and employment around the Lordstown sale and the Stargate announcements to estimate real‑economy multipliers from AI capex. Use FRED for power and employment series, company filings for capex, and Goldman’s published capex scenarios as structural priors. (goldmansachs.com) (datacenterdynamics.com) Goldman’s pair of calls—the bank’s gold target for 2026 and its AI‑capex driven semiconductor forecast—tie the financial and industrial worlds together. The numbers are specific: a raised gold target and a projected half‑trillion‑plus in hyperscaler capex that the market can map into chip revenues and factory renovations like Lordstown. (kitco.com) (goldmansachs.com)