Tech CEOs Are Sandbagging AI Risk, and Investors Are Buying It
Microsoft's latest 10-K filing lists 20 risk factors (prior year had 17 risk factors) of AI-related risks. Not two paragraphs buried in the boilerplate section, seventeen pages. Alphabet, Meta, and Amazon filed similar disclosures in 2024 and 2025. The warnings cover everything from model hallucinations and bias litigation to energy grid constraints and geopolitical supply chain disruptions. Then the same executives get on earnings calls and tell analysts that AI spending will "unlock unprecedented value" and that the capex trajectory is "appropriate given the opportunity." Both things can't be fully true.
The contradiction isn't accidental. It's the core move in a regulatory two-step that every public tech company now has to dance. Under securities law in both the United States and Canada, executives face liability if they fail to disclose material risks that later materialize and tank the stock. So the legal department writes the disclosure. But investor relations still has to sell growth, and growth right now means spending billions per quarter on data centres, Nvidia chips, and power contracts at a scale that makes the fibre buildout of the late 1990s look restrained. Microsoft, Alphabet, Meta, and Amazon collectively scaled their capital expenditures dramatically during 2024 and 2025, reaching over $380 billion annually by year-end 2025. Most of that is data centres, Nvidia chips, and power contracts. If executives flag too much risk, the Street reprices the narrative and the funding dries up before the models start paying back.
The result is a disclosure framework that reads like insurance. By listing every catastrophe, hallucination lawsuits, regulatory fines under Canada's pending Artificial Intelligence and Data Act, talent poaching that compresses margins, energy procurement failures, single-source chip dependency, the companies build a legal safe harbour. When one of those risks lands, they can point to page 147 of the 10-K and say they warned you. In the meantime, the message to buy-side analysts is that those are tail risks, manageable with the right oversight, and the revenue inflection point is two quarters out.
The shift from hype to ROI scrutiny
Investors are starting to notice the gap. Seventy percent of S&P 500 companies mentioned AI in Q1 2024 earnings calls, a record. But by mid-2025, the questions changed. Analysts stopped asking what models companies were deploying and started asking when the productivity gains would show up in the bottom line. The term making the rounds now is "AI utility", can the technology make the company more money, not just make employees faster. That distinction matters. A tool that cuts a process from four hours to two is useful. A tool that cuts a process from four hours to two but costs more to run than the labour it replaced is a cost centre with good PR.
Energy is the risk getting the most airtime in Ottawa and other regulatory centres. The International Energy Agency estimates data centres could double their electricity consumption by 2026. For companies with Net Zero commitments, Microsoft has pledged to be carbon negative by 2030, that creates a bind. You can't scale AI training runs and hit those targets without either buying offsets at a cost that hits margins or delaying the model rollout until green power comes online. Neither option shows up in the bullish analyst decks.
The talent line item is quietly becoming material. Hiring specialized AI engineers in hubs like Toronto and Waterloo now runs in the range of CAD $120,000 to $160,000 in total comp for mid-level roles. That's margin compression dressed up as an investment in the future. The 10-K discloses it. The earnings call doesn't dwell on it.
The sandbagging works because both sides benefit. Executives get to comply with disclosure rules and keep the capital flowing. Investors get to price in the upside and treat the risk section as a legal artifact they can revisit later if something breaks. The only group that loses is the one holding the stock when the power grid can't keep up or when regulators in the EU or Canada decide the risk disclosures weren't specific enough and levy fines that weren't modelled into the guidance. By then, the warnings were already on file. Nobody can say they weren't told.
Microsoft's latest 10-K filing lists 20 risk factors (prior year had 17 risk factors) of AI-related risks. Not two paragraphs buried in the boilerplate section, seventeen pages. Alphabet, Meta, and Amazon filed similar disclosures in 2024 and 2025. The warnings cover everything from model hallucinations and bias litigation to energy grid constraints and geopolitical supply chain disruptions. Then the same executives get on earnings calls and tell analysts that AI spending will "unlock unprecedented value" and that the capex trajectory is "appropriate given the opportunity." Both things can't be fully true.
The contradiction isn't accidental. It's the core move in a regulatory two-step that every public tech company now has to dance. Under securities law in both the United States and Canada, executives face liability if they fail to disclose material risks that later materialize and tank the stock. So the legal department writes the disclosure. But investor relations still has to sell growth, and growth right now means spending billions per quarter on data centres, Nvidia chips, and power contracts at a scale that makes the fibre buildout of the late 1990s look restrained. Microsoft, Alphabet, Meta, and Amazon collectively scaled their capital expenditures dramatically during 2024 and 2025, reaching over $380 billion annually by year-end 2025. Most of that is data centres, Nvidia chips, and power contracts. If executives flag too much risk, the Street reprices the narrative and the funding dries up before the models start paying back.
The result is a disclosure framework that reads like insurance. By listing every catastrophe, hallucination lawsuits, regulatory fines under Canada's pending Artificial Intelligence and Data Act, talent poaching that compresses margins, energy procurement failures, single-source chip dependency, the companies build a legal safe harbour. When one of those risks lands, they can point to page 147 of the 10-K and say they warned you. In the meantime, the message to buy-side analysts is that those are tail risks, manageable with the right oversight, and the revenue inflection point is two quarters out.
The shift from hype to ROI scrutiny
Investors are starting to notice the gap. Seventy percent of S&P 500 companies mentioned AI in Q1 2024 earnings calls, a record. But by mid-2025, the questions changed. Analysts stopped asking what models companies were deploying and started asking when the productivity gains would show up in the bottom line. The term making the rounds now is "AI utility", can the technology make the company more money, not just make employees faster. That distinction matters. A tool that cuts a process from four hours to two is useful. A tool that cuts a process from four hours to two but costs more to run than the labour it replaced is a cost centre with good PR.
Energy is the risk getting the most airtime in Ottawa and other regulatory centres. The International Energy Agency estimates data centres could double their electricity consumption by 2026. For companies with Net Zero commitments, Microsoft has pledged to be carbon negative by 2030, that creates a bind. You can't scale AI training runs and hit those targets without either buying offsets at a cost that hits margins or delaying the model rollout until green power comes online. Neither option shows up in the bullish analyst decks.
The talent line item is quietly becoming material. Hiring specialized AI engineers in hubs like Toronto and Waterloo now runs in the range of CAD $120,000 to $160,000 in total comp for mid-level roles. That's margin compression dressed up as an investment in the future. The 10-K discloses it. The earnings call doesn't dwell on it.
The sandbagging works because both sides benefit. Executives get to comply with disclosure rules and keep the capital flowing. Investors get to price in the upside and treat the risk section as a legal artifact they can revisit later if something breaks. The only group that loses is the one holding the stock when the power grid can't keep up or when regulators in the EU or Canada decide the risk disclosures weren't specific enough and levy fines that weren't modelled into the guidance. By then, the warnings were already on file. Nobody can say they weren't told.
Sources
Read Next
How Dual Citizens Can Claim RESP Tax Benefits Without Form 3520 Reporting
Bond Markets Are Pricing In Recovery, Not the 1970s Replay Already Underway
Why Fortress Tells Private Credit Lenders to Stop Chasing AI Data Centre Deals
Canada's Tax Code Punishes Work and Rewards Wealth Hoarding: Four Reforms That Would Actually Fix It