The 50-year-old law that governed every software company just broke. Here’s what replaces it

The article argues that AI has effectively repealed Brooks’s Law by allowing companies to scale output with smaller teams and heavier compute investment, with internal data cited showing large AI firms generate nearly 3x the revenue run rate per employee versus non-AI software peers. It highlights OpenAI, Anthropic, and Cursor as examples of companies growing from a few million dollars of revenue to billions in under two years. The piece is commentary rather than a company-specific event, so direct market impact is limited.

Analysis

The investable implication is not that “AI is good for software,” but that the production function for certain software businesses is changing from labor-limited to capital-limited. That should compress the gap between winners and fast followers in model-heavy categories, while widening the gap between companies with access to low-cost compute and everyone else. In private markets, that favors firms with balance-sheet endurance and distribution leverage more than pure engineering talent; in public markets, it likely rerates the few beneficiaries that can convert capex into recurring usage before competition arbitrages away pricing power.

Second-order effects are more important than the headline productivity story. If smaller teams can ship faster, procurement cycles shorten and incumbent enterprise vendors face faster feature obsolescence, which could pressure legacy software multiples even before revenue decelerates. Hardware and infrastructure providers are the clearest near-term toll collectors, but their upside is more cyclical than structural: once model training becomes a scale game, the marginal dollar of spend can migrate from experimental software headcount to datacenter capacity, networking, power, and inference optimization.

The contrarian risk is that investors are extrapolating the current scarcity of AI winners into a permanent regime. Over the next 6–18 months, the more likely failure mode is not that AI productivity disappoints, but that returns concentrate so heavily in a small set of model/platform names that downstream app companies see margin compression from rising compute costs and aggressive pricing by the model layer. That makes the broad “AI software” basket vulnerable to dispersion: the market may be paying for an across-the-board uplift when the actual outcome is a barbell.

For IBM specifically, the article is not a near-term earnings catalyst; the more relevant read-through is competitive: legacy workflow vendors with large installed bases may benefit only if they can bundle AI into existing distribution, but they also face the risk of becoming the distribution layer for someone else’s compute economics. The market should treat this as a multi-quarter reallocation story, not a one-month momentum trade.

Key Decisions for Investors

• Long MSFT / NVDA as the cleanest public-market expression of 'capital-limited AI' over a 6-12 month horizon; MSFT monetizes distribution while NVDA captures the compute tollbooth. Use pullbacks to 20-day moving averages; downside risk is multiple compression if enterprise spend slows.
• Short a basket of legacy enterprise software names with high labor intensity and weak AI differentiation (e.g., CRM/ORCL-style exposure) vs long a model/platform leader; hold 3-9 months. Thesis: AI accelerates feature parity and shortens replacement cycles, pressuring subscription pricing before it boosts volume.
• In private-markets exposure, favor late-stage AI infrastructure and workflow winners over broad software venture funds for the next 12-24 months. The trade is less about upside magnitude than surviving the capital race; back firms that can absorb large checks without forcing premature dilution.
• Use a pair trade: long semiconductor/electrical infrastructure beneficiaries, short high-multiple application software indices. Risk/reward is attractive if AI capex remains elevated for 2-4 quarters; exit if capex guidance rolls over or inference economics deteriorate faster than expected.
• Do not add to IBM on this thesis alone; treat it as a potential relative loser if it cannot defend distribution economics against AI-native competitors. Only revisit if evidence emerges that AI attach rates improve gross margin without higher services intensity.