How much of the Generative AI boom is real value, and how much is simply a financial mirage? A recent MIT‑linked report argues that despite $30–$40bn of enterprise spending, roughly 95% of organisations see zero return from GenAI initiatives (NANDA: The GenAI Divide). Other outlets have picked up the same stark finding (see The Register and Fortune) and, perhaps my favourite, Ed Zitron.
1 — Why GenAI looks and behaves like a bubble
There are three signals you should recognise: feverish capital inflows, sky‑high valuations for product‑thin startups, and a proliferation of undifferentiated point solutions. Venture capital and corporate R&D have rushed toward anything labelled “GenAI” because the market attention feels like a moat. That creates perverse incentives: build fast, ship a demo, raise the next round — not necessarily solve a real user problem.
From a product perspective, many projects shortcut discovery. Teams buy APIs from providers such as OpenAI, Anthropic or integrate large models via platforms like Microsoft or DeepMind, then retrofit a use case. The result is a cosmetic feature rather than a sustained capability: it looks like innovation, but it often lacks defensibility, measurable outcomes and the data backbone that creates compounding value.
2 — The real cost: opportunity lost for critical sectors
Capital is not neutral. When tens of billions chase the next generative trick, it diverts investment away from long‑term, hard problems: education, climate tech, public infrastructure, and core data engineering. These are areas that require patient capital and systems thinking rather than rapid productisation.
Take education: product leaders know that improving learning outcomes demands deep pedagogy, longitudinal trials and teacher training — not just a “chatty tutor” widget. Similarly, environmental systems need robust IoT, sensors and domain modelling, which rarely deliver immediate headline metrics attractive to short‑term investors. If the market funnels scarce talent and funding into shiny GenAI demos, those structural investments stall and society pays the price.
3 — Not all GenAI is worthless — where value really appears
It’s important to be clear: some organisations do extract enormous value. The MIT work itself shows a sharp divide: a small proportion of teams capture meaningful returns by combining models with deep product and domain expertise. What separates the winners?
- Data flywheels: firms that invest in high‑quality, proprietary data and instrumentation turn one‑off prompts into iterative, improving experiences.
- Workflow embedding: those who integrate models into existing, high‑value processes (e.g., legal review, clinical decision support, financial desk tools) drive measurable time or cost savings.
- Regulatory and operational rigor: regulated sectors that treat AI as a long game — with governance, monitoring and human‑in‑the‑loop controls — avoid the puff pieces and achieve durable impact.
Examples are instructive. Large incumbents that combine deep domain teams with model capabilities — for instance, Microsoft’s investments across its productivity stack or Google’s integration of models into search and cloud services — are more likely to see reliable returns than small vendors selling isolated prompts.
4 — What product leaders should do today
If you’re a CPO, CTO or CEO, treat Generative AI like another tool in the toolbox — powerful, but not a panacea. Practical steps:
- Start with outcome metrics: instrument your pilots to measure customer value, not tech metrics. Time saved, conversion uplift, retention — these are what pay the bills.
- Protect strategic capital: ring‑fence investment for long‑term bets in education, climate and core engineering. Don’t let marketing narratives decide R&D allocation.
- Invest in data & integration: models are commoditised quickly. Your moat is data quality, feedback loops and trusted integrations with business processes.
- Adopt staged experimentation: run small, properly powered trials with clear go/no‑go criteria. Scale when you see reproducible impact.
- Build people capability: reskilling and change management are more important than a model subscription. Outcomes depend on humans designing, curating and governing systems.
5 — A constructive way forward
Bubbles and hype cycles are part of technological progress. They accelerate exploration, surface novel ideas and sometimes create infrastructure that later benefits other fields. But this time the stakes are high: misallocated capital hurts not only investors but public goods and future generations. Product leaders must be disciplined custodians of scarce resources.
That means saying “no” to vanity AI projects, demanding ROI, and steering teams toward integrations that are measurable, defensible and aligned with user value. It also means lobbying boards and investors for balanced portfolios where patient capital funds the slow, hard work — in education, energy transition and public services — that will deliver the biggest societal returns.
Generative AI is not a fad, but the current market behaviour looks dangerously bubble‑like. The useful technologies born today will be the foundation for tomorrow’s products — provided we don’t let speculation drown out sustainable investment. If you lead product or technology, your most important decision right now is not which model to use, but where you choose to place your bets.
Leave a Reply