As Moore's law goes to computing, so does AI. Here’s What That Means for Business
Stanford’s latest AI Index reveals how fast things are moving — and where leaders need to catch up
The Stanford AI Index 2025 is out, and the signal is clear: artificial intelligence is no longer a research tool — it’s infrastructure. AI systems are faster, cheaper, and more capable than ever. Enterprise adoption has surged. But so have the risks, from brittle outputs to mounting real-world incidents.
Here’s what the newest data tells us — and how smart companies are adjusting strategy.
AI Adoption Has Gone Mainstream — and Matured
AI is now deeply embedded across sectors. According to Stanford’s new data, 78% of global organizations are using AI in production — up from 55% in 2023.
At Morgan Stanley, over 98% of advisor teams use GPT-powered tools to draft reports and client recommendations. In banking, retail, and logistics, companies are moving beyond experimentation to make AI a standard operating layer. Think forecasting, fraud detection, and generative assistants built into everyday workflows.
The key driver? Foundation models are now widely accessible via APIs, removing the need for in-house AI expertise.
The operational insight:
Leaders no longer have the luxury of treating AI as “experimental.” It’s already a performance differentiator. The question is whether your organization is using it deliberately — with measurable impact.
AI Costs Are Plummeting — and the Gap Is Closing
AI performance keeps rising — but costs are falling fast.
Training a GPT-3.5-level model today is 280× cheaper than just a few years ago. Open-source challengers like Mistral, MosaicML, and LLaMA are cutting the cost of large models dramatically, and cloud providers are making inference even more affordable.
For example, MosaicML trained GPT-3-level models for under $1M, compared to OpenAI’s original $4M+ cost. Meanwhile, NVIDIA’s Blackwell GPUs and Google’s next-gen TPUs are making inference pricing plummet — unlocking real-time AI services at scale.
The operational insight:
AI is no longer a big-tech advantage. The barrier to entry is gone. If you’re still calculating ROI based on 2020 pricing, you’re missing opportunities that are now viable — and competitors are already exploring them.
Safety Incidents Are Spiking — and Everyone’s at Risk
More adoption means more exposure. The Stanford report highlights a troubling trend: AI-related incidents have increased 25× over the last decade, with a sharp spike in 2024–2025.
Recent cases include:
- An autonomous driving system failing to recognize a pedestrian
- A mental health chatbot offering harmful advice
- Biased algorithms producing discriminatory outcomes
- Large language models hallucinating critical information
These aren’t edge cases — they’re systemic indicators that even well-trained models can behave unpredictably when deployed at scale. Generative AI’s tendency to “hallucinate” confident but false information is now one of the top safety concerns flagged by enterprise risk teams in 2025. Stanford tracked dozens of high-profile failures in just the last 12 months, many involving generative systems deployed without guardrails.
The operational insight:
The operational insight: If you’re deploying AI, safety isn’t a checkbox — it’s core to brand trust. Governance frameworks, rigorous model audits, and human-in-the-loop review are essential, but not sufficient. Enterprises need to expand QA processes to include structured prompt testing and red-teaming — systematically probing models for edge cases, bias triggers, and failure patterns before exposure to users. Teams like OpenAI and Meta now conduct multi-phase prompt audits across regions and personas — a standard that’s quickly becoming table stakes.
Policy Is Still Catching Up — And You Can’t Wait
Despite the headlines, legislation hasn’t caught up with AI’s pace. The U.S. proposed 221 AI-related laws in 2024, but only 4 passed. The EU’s AI Act is still not in effect.
Most of what exists are executive orders or voluntary standards — like the U.S. AI Safety Executive Order, which encourages model testing but doesn’t mandate it.
The operational insight:
The regulatory gap puts pressure on companies to self-govern — but that’s also an opportunity. Organizations that adopt proactive frameworks (like NIST’s AI Risk Framework or OECD AI Principles) are better positioned to win trust, attract partners, and avoid future compliance headaches.
AI Is Accelerating — With or Without You
Stanford’s 2025 Index leaves no doubt: AI is advancing on all fronts. Adoption is broad. Costs are falling. Risks are growing. And regulation is lagging behind.
The winners in this new environment will be the companies that move fast — but not blindly.
AI is no longer a tech initiative. It’s a strategic advantage.
Make it usable. Make it safe. Make it count.