Odd Lots: Goldman’s Argenti on the Improvements in AI (Podcast)

Goldman Sachs CIO Marco Argenti says the bank is accelerating AI deployment, moving from building internal tools to integrating agentic platforms (e.g., Claude Code) and using AI-assisted coding that is materially changing developers' workflows. The firm is focused on addressing substantial data, integration and regulatory/compliance challenges as it scales AI across the organization since the last update 18 months ago.

Analysis

Goldman’s internal push into agentic AI creates an asymmetric opportunity: software engineering output can jump materially (we model a 20–40% effective productivity lift for development cycles within 6–12 months), but that does not translate 1:1 into net income — most of the early value accrues to faster feature releases and lower marginal FTE spend, while gross compute and licensing costs rise. Expect bank-level operating leverage to improve only after 12–24 months as pilots move into production and amortizable software assets build up; in the interim, elevated cloud/GPU bills and compliance staffing will compress reported margins.

Second-order winners are infrastructure and governance vendors — cloud providers, GPU suppliers and specialized data lineage/cybersecurity firms — which capture recurring revenue as banks operationalize agentic workflows. Smaller fintechs and boutique trading shops are disadvantaged: they face a higher fixed-cost hurdle to match Goldman's integrated stack and auditability, raising the threshold for profitable scale and making them prime targets for customer attrition or acquisition in the next 12–36 months.

Key risks are regulatory and operational rather than purely technical: a major audit failure, data leakage, or an enforcement action around model explainability could produce a near-term re-rating within 30–90 days and force on-prem or restricted-cloud deployments that materially increase capex. Conversely, a successful evidence package to regulators (proof of lineage, red-teaming results) would be a catalyst that unlocks commercial rollout and outsized margin capture over 12–24 months. Contrarian read: the market is overstating immediate revenue upside from AI for banks but understating the duration and stickiness of incremental infrastructure spend — infrastructure/vendors likely harvest most early dollars while front-office P&L improvements lag.

Key Decisions for Investors

• GS — Tactical long: buy GS 6–12 month call options (target 3:1 upside vs premium) or a modest outright overweight in the equity. Rationale: optionality to deliver margin expansion when production deployments scale; hedge with a 25–35% notional put protection to cap downside from regulatory headlines over 90 days.
• NVDA — Leveraged long: buy 3–6 month calls (or call spreads) to capture near-term GPU demand from banks and buy-side scaling agentic platforms. Risk/reward: high gamma; losses limited to premium, upside substantial if enterprise deployments accelerate over 3–6 months.
• CRWD or PANW — Buy 6–12 month calls on cybersecurity/data-governance vendors to play increased spend on monitoring, lineage and red-teaming. Expect steady recurring revenue pickup; downside is muted relative to pure play software if budgets are reallocated.
• Pair trade — Long GS equity / Short a smaller fintech (e.g., SQ) equal notional for 3–12 months: longs capture integrated-platform advantages and compliance moat; shorts expose smaller firms’ higher execution and regulatory risk. Target a 1.5–2.0 Sharpe improvement versus outright GS exposure; monitor regulatory developments weekly and trim if enforcement risk rises.