Lari Hämäläinen

AI analyst and strategist at QuantumBlack, AI by McKinsey.

Mentions

Appeared in

Discussed in

Key positions and views

The vast majority of enterprises are failing at AI implementation, stuck in a cycle of high investment, low ROI, and high project cancellation rates.

The capability of the underlying AI model is the primary driver of value, as shown by the dramatic turnaround in developer productivity being caused solely by the release of new models.

The AI industry is consolidating due to the exponentially growing compute costs of training frontier models, which is simultaneously leading to diminishing performance differences between top models.

AI will fundamentally reshape the labor market by automating up to 68% of white-collar jobs, which will dramatically increase skill inequality and the need for mass adult reskilling.

Standardized LLM benchmarks have become saturated and are no longer meaningful for differentiating frontier models; real-world, complex task automation is the new measure of performance.

Podcast consensus on Hämäläinen

Points of consensus

▶Enterprise AI adoption is broadly failing to deliver business value, with 95% of McKinsey clients reporting no P&L impact and only 1% of enterprises having mature, value-generating deployments.Apr 2026

▶The rate of enterprise AI initiative failure is high and increasing, with cancellation rates exceeding 40% in October 2025, a 2.5-fold increase from the previous year.Apr 2026

▶The cost and scale of training frontier AI models are escalating dramatically, with compute investment growing 4.5x year-over-year, leading to a projected $7-8 trillion in data center investments by 2030.Apr 2026

▶Developer productivity with AI tools experienced a dramatic reversal from a 20% decline to a 20% gain, driven almost entirely by the release of more capable models like Gemini 3 and Claude Opus 4.5 in late 2025.Apr 2026

Points of debate

▶Hämäläinen highlights a significant tension between the massive, multi-trillion dollar investments being poured into AI infrastructure and the near-total lack of tangible P&L impact for the vast majority of enterprises.

▶There is a contrast between the exponential growth in compute power for training models and the diminishing returns in performance differences between competing frontier models, suggesting a saturation of standard benchmarks.Apr 2026

▶A paradox exists between AI's immense potential for automation (up to 68% of white-collar jobs) and the high cancellation rate (over 40%) of corporate AI initiatives, indicating a major gap between technological capability and successful implementation.Apr 2026

▶Hämäläinen's narrative presents a volatile 'productivity paradox' where initial AI tools actively harmed developer productivity before a subsequent generation of models created significant gains, challenging the idea of a smooth adoption curve.Apr 2026

Key themes

▶The Enterprise AI 'Value Gap'Apr 2026

Hämäläinen consistently highlights a chasm between the immense investment in AI and its failure to produce tangible business results. He claims 95% of enterprises see no P&L impact, only 1% have mature deployments, and over 40% of initiatives were canceled in late 2025, painting a picture of widespread disillusionment.

For investors, this theme suggests that the primary challenge in AI is no longer technology access but organizational change, strategic alignment, and the ability to translate AI capabilities into specific business workflows, making implementation expertise a key differentiator.

▶AI-Driven Developer Productivity RevolutionApr 2026

This theme traces the dramatic turnaround in AI's impact on software development. Hämäläinen notes that early tools caused a 20% productivity decline, but the release of models like Gemini 3 and Claude 4.5/4.6 in late 2025 flipped this to a 20% productivity gain and a sevenfold increase in AI-generated code commits.

This indicates that the utility of AI tools is almost entirely dependent on the raw capability of the underlying model, suggesting that breakthroughs in enterprise value will be tightly coupled to frontier model release cycles rather than incremental software improvements.

▶The Sobering Economics of Frontier AIApr 2026

Hämäläinen emphasizes the escalating resource requirements for creating new AI models. He points to a 4.5x year-over-year increase in training compute and projects a staggering $7-8 trillion in data center investments by 2030, which is causing the number of new frontier models to decrease.

This trend points toward a future market dominated by a few hyper-capitalized players, making access to proprietary, state-of-the-art models a significant competitive advantage and a potential systemic risk.

▶AI's Looming Impact on Labor and SkillsApr 2026

Hämäläinen forecasts a massive societal disruption driven by AI, predicting it could automate 44% of all US jobs (68% of white-collar roles). This will necessitate a 3-5x increase in adult reskilling and is expected to cause a 3-4x growth in skill inequality.

Analysts should monitor labor market data for signs of this disruption, as it implies significant long-term shifts in workforce demand, wage structures, and the value of different skills, creating both risks for exposed industries and opportunities in education and reskilling.

Source episodes

Sentiment over time

Not enough data for timeline

Changes over time

May 2025

Cites a METR benchmark study showing early AI coding tools led to a surprising 20% decline in developer productivity, contrary to expectations of a 30% gain.

October 2025

Describes a critical inflection point where over 40% of enterprise AI initiatives were canceled, while simultaneously, the release of Gemini 3 and Claude Opus 4.5 began to reverse the negative productivity trend.

Post-October 2025

Notes a sevenfold increase in the volume of AI-generated code committed to GitHub via Claude Code, indicating rapid adoption following new model releases.

January 2026

References a follow-up study confirming the productivity shift, with AI tools now delivering an absolute productivity gain of 20% for developers.

End of 2026 (Prediction)

Predicts that LLMs will achieve human-level proficiency in using enterprise tools and PCs, and that 20% of all code written globally will be AI-generated.

End of 2030 (Prediction)

Projects that cumulative data center investments for AI will reach $7 to $8 trillion.

Suggested prompts

How does Hämäläinen reconcile the massive projected investment in AI infrastructure with the widespread lack of P&L impact reported by enterprises? &nearr;What specific organizational or strategic changes must accompany the adoption of models like Claude 4.6 Opus to translate developer productivity gains into enterprise-wide value? &nearr;Based on Hämäläinen's claims, what are the leading indicators that an enterprise AI initiative is likely to be canceled? &nearr;What are the second-order effects of Hämäläinen's prediction that skill inequality will grow by three to four times due to AI adoption? &nearr;

Key concepts

Enterprise AI Adoption 1 ep Developer Productivity 1 ep AI Investment & ROI 1 ep Labor Market Disruption 1 ep Frontier AI Models 1 ep Compute Requirements 1 ep AI-Generated Code 1 ep Skill Inequality 1 ep LLM Benchmarks 1 ep

Notable quotes

“95% of McKinsey's clientele are saying that we don't see P&L impact.”

Lari Hämäläinen · Building the Enterprise of the Future presented by QuantumBlack, AI by McKinsey

“Their expectation was about 30% productivity uplift. In real life, developer productivity declined about 20%.”

Lari Hämäläinen · Building the Enterprise of the Future presented by QuantumBlack, AI by McKinsey

“We got Gemini in end of October, and we got Gemini 3, and we got Claude Opus 4.5 in October. That was the single factor that drove the change.”

Lari Hämäläinen · Building the Enterprise of the Future presented by QuantumBlack, AI by McKinsey

“if you look at the data center investments going into AI, we're talking of roughly $7 to $8 trillion by end of 2030.”

Lari Hämäläinen · Building the Enterprise of the Future presented by QuantumBlack, AI by McKinsey

Report last updated: Apr 21, 2026

Get started free

Back to Entities Intelligence Report