What are the key takeaways from this Masters of Scale episode?

Key insights include: **AI Model Commoditization:** Foundation models will become commodities within two to three years, meaning switching costs between providers drop near zero. As GPU pricing has already doubled in six months, enterprises should begin optimizing which model size handles which task rather than defaulting to large frontier models for every workload, reducing token costs significantly.; **AI Implementation ROI Timeline:** Expect negative returns for the first six to twelve months of AI deployment. IBM spent more than it saved initially due to engineer costs, infrastructure, and opportunity costs. Returns turned 10x positive after year two, reaching $4.5 billion in savings against 2022 baseline spending by year four — scale is the prerequisite.; **Right-Sizing AI Tools:** Enterprises currently use large frontier models for tasks that smaller, cheaper, on-premise models handle adequately — analogous to using an 18-wheeler for grocery runs. IBM predicts this mismatch corrects within 24 months as token prices rise, forcing cost-conscious procurement of fit-for-purpose models rather than one-size-fits-all deployments.

What did Arvind Krishna discuss on Masters of Scale?

IBM CEO Arvind Krishna outlines why foundation models will become commodities within two to three years, why enterprises are mismatching AI tools to tasks, how IBM extracted $4.5 billion in efficiency gains from AI deployment, and why the company is betting $10 billion on quantum computing as the next computing frontier. Key topics include: **AI Model Commoditization:** Foundation models will become commodities within two to three years, meaning switching costs between providers drop near zero. As GPU pricing has already doubled in six months, enterprises should begin optimizing which model size handles which task rather than defaulting to large frontier models for every workload, reducing token costs significantly.; **AI Implementation ROI Timeline:** Expect negative returns for the first six to twelve months of AI deployment. IBM spent more than it saved initially due to engineer costs, infrastructure, and opportunity costs. Returns turned 10x positive after year two, reaching $4.5 billion in savings against 2022 baseline spending by year four — scale is the prerequisite..

How long is this episode of Masters of Scale?

This episode is 41 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

Masters of Scale

IBM’s $10 billion bet on what comes after AI

June 18, 2026

41 min episode · 2 min read

Arvind Krishna

Episode

41 min

Read time

2 min

Topics

Productivity, Fundraising & VC, Leadership

AI-Generated Summary

Published Jun 18, 2026

Key Takeaways

✓AI Model Commoditization: Foundation models will become commodities within two to three years, meaning switching costs between providers drop near zero. As GPU pricing has already doubled in six months, enterprises should begin optimizing which model size handles which task rather than defaulting to large frontier models for every workload, reducing token costs significantly.
✓AI Implementation ROI Timeline: Expect negative returns for the first six to twelve months of AI deployment. IBM spent more than it saved initially due to engineer costs, infrastructure, and opportunity costs. Returns turned 10x positive after year two, reaching $4.5 billion in savings against 2022 baseline spending by year four — scale is the prerequisite.
✓Right-Sizing AI Tools: Enterprises currently use large frontier models for tasks that smaller, cheaper, on-premise models handle adequately — analogous to using an 18-wheeler for grocery runs. IBM predicts this mismatch corrects within 24 months as token prices rise, forcing cost-conscious procurement of fit-for-purpose models rather than one-size-fits-all deployments.
✓AI Adoption Strategy — Focus Over Breadth: Rather than running 100 AI experiments simultaneously, Krishna recommends selecting three to five use cases and deploying them fully at scale. This teaches change management, data organization, and process redesign. Once that playbook is proven, expand to 10, then 20 initiatives — building organizational confidence incrementally rather than spreading resources thin.
✓Quantum Computing Acceleration: IBM's quantum systems progressed from simulating 5-atom molecules in summer 2025 to 12,000 atoms by April — approaching the protein simulation range of 10,000–40,000 atoms. Enterprises should begin developing quantum algorithms now so they are deployment-ready when hardware matures, treating quantum preparation as parallel work alongside AI, not a future-state decision.

What It Covers

IBM CEO Arvind Krishna outlines why foundation models will become commodities within two to three years, why enterprises are mismatching AI tools to tasks, how IBM extracted $4.5 billion in efficiency gains from AI deployment, and why the company is betting $10 billion on quantum computing as the next computing frontier.

Key Questions Answered

•AI Model Commoditization: Foundation models will become commodities within two to three years, meaning switching costs between providers drop near zero. As GPU pricing has already doubled in six months, enterprises should begin optimizing which model size handles which task rather than defaulting to large frontier models for every workload, reducing token costs significantly.
•AI Implementation ROI Timeline: Expect negative returns for the first six to twelve months of AI deployment. IBM spent more than it saved initially due to engineer costs, infrastructure, and opportunity costs. Returns turned 10x positive after year two, reaching $4.5 billion in savings against 2022 baseline spending by year four — scale is the prerequisite.
•Right-Sizing AI Tools: Enterprises currently use large frontier models for tasks that smaller, cheaper, on-premise models handle adequately — analogous to using an 18-wheeler for grocery runs. IBM predicts this mismatch corrects within 24 months as token prices rise, forcing cost-conscious procurement of fit-for-purpose models rather than one-size-fits-all deployments.
•AI Adoption Strategy — Focus Over Breadth: Rather than running 100 AI experiments simultaneously, Krishna recommends selecting three to five use cases and deploying them fully at scale. This teaches change management, data organization, and process redesign. Once that playbook is proven, expand to 10, then 20 initiatives — building organizational confidence incrementally rather than spreading resources thin.
•Quantum Computing Acceleration: IBM's quantum systems progressed from simulating 5-atom molecules in summer 2025 to 12,000 atoms by April — approaching the protein simulation range of 10,000–40,000 atoms. Enterprises should begin developing quantum algorithms now so they are deployment-ready when hardware matures, treating quantum preparation as parallel work alongside AI, not a future-state decision.

Notable Moment

Krishna argued that zero risk-taking is actually the highest-risk corporate strategy. Companies that avoid innovation allow competitors to clone their profitable segments, shrinking margins until leadership cuts investment further — accelerating decline toward acquisition or collapse within roughly a decade of the initial conservative pivot.

Know someone who'd find this useful?