Companies promised AI gains— here's how the payoff showed up in earnings

Companies are beginning to quantify AI’s financial returns: Santander attributed €194 million ($216 million) in business value to AI initiatives in the first nine months of 2026. Novartis said AI began tangibly improving its P&L as Q3 core operating income rose 10% year over year at constant currencies, while Standard Chartered reported 14% operating-income growth excluding notable items and cited AI-enabled efficiency. Meta said AI added substantial advertising revenue as quarterly revenue rose 26%; Google Cloud revenue more than doubled to nearly $31 billion, while Carlsberg said AI benefits were not yet at scale.

Analysis

The investable signal is not that AI is “working” broadly; it is that the evidence quality now differs by business model. Meta’s claimed incremental ad revenue is closest to monetization, while Alphabet’s Cloud growth supports demand for AI infrastructure but does not establish attractive returns on the capital required to supply it. Banking and pharma claims remain harder to isolate from normal operating improvement; Santander’s quantified benefit is useful, but company-defined “business value” is not equivalent to recurring savings or earnings. Carlsberg’s lack of scaled impact is a reminder that deployment and measurable P&L leverage have different timelines.

Over the next 1–3 months, earnings calls should reward measurable, recurring outcomes—not AI mentions. Track ad yield/conversion at Meta, Cloud margins and capex at Alphabet, and expense ratios/productivity at Standard Chartered and Santander. For Novartis, the longer-dated upside is faster R&D productivity, but no quantified evidence here supports underwriting that today. Over 6–18 months, broad AI infrastructure spending may benefit suppliers such as Nvidia, while raising depreciation and utilization risk for hyperscalers; spend growth alone is not proof of attractive returns.

Contrarian view: attribution may be overstated because management can relabel ordinary digitization or operating leverage as AI benefit. The market may also underprice the lag between efficiency gains and realized cost reductions, especially where labor, compliance, and process redesign constrain capture. No valuation, stock-reaction, or comparable ROI data are supplied, so avoid treating these statements as a sector-wide earnings upgrade.

Key Decisions for Investors

• No broad AI trade on this evidence. For the next earnings cycle, favor companies that disclose recurring revenue uplift or realized expense savings over those citing deployment, usage, or “business value” without a bridge to reported earnings.
• Watchlist—not an immediate recommendation: consider a relative long META versus a less-proven AI monetization exposure only if subsequent reports sustain ad monetization and show returns keeping pace with investment. Reassess if ad growth weakens or capex rises without better revenue productivity.
• For Alphabet, monitor Cloud operating margin alongside growth and infrastructure spending over the next 1–3 months; strong revenue without margin progress would weaken the inference that AI demand is producing attractive returns. Do not infer Microsoft or Amazon ROI from sector spending expectations.
• Treat bank and pharma AI claims as alerts, not earnings estimates: verify Santander’s definition and repeatability of business value, Standard Chartered’s expense/income trajectory, and Novartis’s separately quantified R&D productivity before changing positions. Carlsberg is a useful control case; revisit only if management reports scaled, measurable savings.
• Falsification: reduce the positive AI-ROI thesis if the next 1–2 reporting cycles show no measurable productivity or margin improvement despite continued investment, or if infrastructure costs outpace monetization. Key missing data include company-level AI capex, recurring savings, attribution methodology, and stock valuations.