The Stanford Institute for Human-Centered Artificial Intelligence (HAI) has released its 2025 AI Index Report, offering a data-driven snapshot of AI’s breakneck progress and the challenges shaping its future.

Since its first benchmark in 2022, AI has emerged as the foremost technological concern of our time – models have grown more powerful, investments have surged, and governments have scrambled to keep pace with regulation.

The report delivers a comprehensive analysis of AI’s technical advancements, economic impact, policy shifts, and ethical concerns, serving as an essential guide to understanding where AI stands today – and where it’s headed next.

This year’s findings reveal record-breaking performance in AI capabilities, skyrocketing private sector funding, and accelerating real-world adoption across industries.

Yet the report also highlights persistent gaps in reasoning, safety, and equitable access – reminders that as AI integrates deeper into society, the stakes for responsible development have never been higher.

At 456 pages long across eight chapters, the report hardly makes for light reading. As such, we’re going to roundup the key takeaways from each chapter.

1 – Research and Development

AI research continues to be shaped by a growing divide between academia and industry – each pushing the field forward, but in distinct ways.

While industry now dominates the creation of notable AI models – with nearly 90% emerging from corporate labs in 2024, up from 60% the year before – academic institutions still lead when it comes to the most influential research. For three years running, universities have been the top producers of highly cited AI publications.

Geopolitically, China remains unmatched in sheer research output, producing nearly a quarter of all AI publications and citations last year. Yet it’s the United States that consistently leads on influence, contributing the top-100-cited papers.

Meanwhile, AI research is expanding across scientific disciplines at breakneck pace. In the last decade, the total number of AI-related publications has more than doubled, with AI now making up over 40% of all computer science papers.

Model development continues to be dominated by the United States. In 2024 alone, American institutions released 40 notable models – more than double China’s total, and far outpacing Europe. But the scale comes with consequences. AI systems are becoming increasingly resource-intensive to build, with training compute doubling roughly every five months and emissions skyrocketing.

The carbon footprint of training today’s most advanced models is now measured in thousands of tons, which is sobering.

Paradoxically, as AI models grow larger and more expensive to train, they’re becoming dramatically cheaper to use. The cost of querying a model equivalent to GPT-3.5 has plummeted more than 280-fold since late 2022.

Hardware improvements have helped fuel this transformation, with chips becoming faster, more energy efficient, and more cost-effective every year. AI patenting has surged alongside this technological boom, led by China, and while concerns about sustainability loom, the momentum behind AI R&D shows no sign of slowing.

2 – Technical Performance

AI systems are improving at a pace few could have predicted.

In just a year, models have leapt forward on difficult new benchmarks like MMMU, GPQA, and SWE-bench – tests designed to push them to their limits. Coding tasks once unsolvable are now handled with ease, with AI systems improving from 4% to over 70% success rates in some cases.

Open-weight models have also seen a remarkable catch-up. The performance gulf between open and closed systems, once wide, has nearly disappeared. Similarly, the once-significant lead held by U.S. models over their Chinese counterparts has shrunk dramatically across multiple benchmarks, reflecting a global evening of capabilities.

At the frontier, distinctions between top-tier models are blurring. The Elo score gap between the best and the tenth-best models is nearly half what it was a year ago, as more players deliver high-performing systems. Some of this edge comes from innovations in how models think – literally. New “test-time compute” methods allow models to reason step-by-step, resulting in far better problem-solving at the expense of speed and cost.

Meanwhile, new benchmarks are pushing boundaries further, often exposing the limits of today’s systems. On Humanity’s Last Exam, for instance, even the best model scored under 9%.

Essentially, AI still struggles with logical reasoning and scalability, especially on tasks that exceed their training scope. And while large-scale models get most of the headlines, 2024 revealed a quiet revolution: smaller models now routinely outperform what once required hundreds of billions of parameters.

The generative space also had a breakout year. Video generation tools like SORA, Movie Gen, and Veo 2 are producing higher-quality results than ever before – though this comes with its own swath of concerns around copyright and creative ethics.

AI agents are also beginning to rival human performance – at least in short tasks – proving faster and more cost-efficient in certain domains.

Still, when complexity or time increases, humans maintain the upper hand. For now.

3 – Responsible AI

Responsible AI remains a pressing concern – and an area where progress is frustratingly slow.

While more organisations are vocal about the risks, fewer are taking concrete action. Cybersecurity, accuracy, and compliance are frequently cited as top concerns, but mitigation efforts remain uneven. New benchmarks like HELM Safety and AIR-Bench are helping to evaluate risks more rigorously, but responsible AI evaluation is still far from standard practice – and is certainly not commensurate with the pace of development and adoption.

The number of AI-related incident reports surged in 2024, hitting an all-time high. Yet as these risks grow more visible, so does global interest in mitigating them. Institutions across continents – from the OECD and the EU to the African Union – have stepped up with frameworks to guide responsible development, focusing on transparency, explainability, and trust.

Still, structural challenges persist.

One of the most troubling is the rapid erosion of the data commons. More and more websites are limiting access to the data that AI models rely on, raising concerns about the diversity and quality of future training sets. Transparency, while improving, also has a long way to go. The latest Foundation Model Transparency Index shows notable progress but leaves plenty of room for growth.

Factuality remains a moving target.

New tools like FACTS and SimpleQA offer hope for better truthfulness benchmarks, though they’ve yet to be universally adopted. And while some models are explicitly trained to reduce bias, implicit biases remain deeply embedded – affecting how AI systems associate race, gender, and expertise.

These persistent problems are drawing increasing attention from the academic community, where research into responsible AI is growing rapidly.

4 – Economy: Investment Booms, Adoption Soars

The AI gold rush is in full swing.

Global private investment surged to $252.3 billion in 2024, an increase of 44.5% from the previous year. The U.S. dominates this spending, outpacing China 12-to-1 in total AI investment and holding a $25.4 billion lead in generative AI funding.

Adoption is also skyrocketing: 78% of organisations now use AI, up from 55% in 2023, while generative AI tools have infiltrated 71% of businesses – more than double last year’s figure.

However, tangible financial gains remain modest. While 49% of firms report cost savings in service operations and 71% see revenue bumps in marketing, most benefits hover below 5–10%.

China’s industrial robotics stronghold persists, installing 51% of global units (276,300 in 2023), though collaborative robots – now 10.5% of new installations – signal a shift toward human-AI teamwork.

Meanwhile, Big Tech is betting big on energy: Microsoft’s $1.6 billion nuclear deal to power AI data centres underscores the sector’s voracious appetite. The payoff? Studies confirm AI boosts productivity and narrows skill gaps, even as costs plummet.

5 – Science & Medicine: AI Rewrites the Rules

AI is no longer just an assistant in the lab – it’s a Nobel laureate.

AlphaFold’s creators clinched the 2024 Chemistry Prize for revolutionizing protein folding, while neural network pioneers Hinton and Hopfield took Physics honors. Their legacy lives on in models like ESM3 and AlphaFold 3, whose precision grows with scale, and Med-Gemini, a foundation model now acing 96% of MedQA clinical benchmarks – outperforming doctors in complex diagnoses.

The FDA’s stamp of approval tells the same story: AI-enabled medical devices skyrocketed from 6 in 2015 to 223 by 2023, with synthetic data unlocking breakthroughs in drug discovery and privacy-safe diagnostics.

Yet challenges linger.

While medical AI ethics papers quadrupled since 2020, protein databases like UniProt expanded 31%, and wildfire predicting FireSat debuted, benchmarks like MedQA near saturation – pushing researchers toward tougher tests.

6 – Policy & Governance: Regulation Catches Up

As AI’s footprint widens, so does lawmakers’ scrutiny. U.S. states passed 131 AI laws in 2024 – more than double 2023’s tally – while federal agencies introduced 59 regulations, up from 25. Deepfake rules spread fastest: 24 states now restrict election-related synthetic media, with 15 joining the fray this year alone.

Globally, legislative mentions of AI rose 21%, as nations raced to secure infrastructure advantage.

7 – Education: Demand Outpaces Preparedness

AI’s classroom invasion is uneven.

While two-thirds of nations now offer K–12 computer science (up from half in 2019), African schools struggle with basic electricity access. In the U.S., master’s degrees in AI nearly doubled in 2023, cementing its lead in tech graduates – yet fewer than half of high school CS teachers feel equipped to teach AI, despite 81% advocating for its inclusion.

The gap reflects a broader divide. While Spain, Brazil, and the U.K. trail the U.S. in producing ICT grads, Turkey shines in gender parity. For now, upskilling remains patchwork – a bottleneck for the next generation of builders.

8 – Public Opinion: Hope Tempered by Fear

Globally, 55% now see AI’s benefits outweighing risks, up 3 points since 2022 – but optimism fractures along regional lines.

Enthusiasm runs highest in China (83%) and Indonesia (80%), while scepticism lingers in the U.S. (39%) and Canada (40%).

Trust is eroding too: just 47% believe AI firms protect data (down from 50%), and 61% of Americans fear self-driving cars, though resistance has eased from 2023’s peak.

Workers brace for disruption: 60% expect AI to reshape their jobs, though only 36% fear replacement.

Policymakers echo the ambivalence – 74% back regulation, with stark partisan splits. Priorities like data privacy (80%) and retraining (76%) draw consensus, but proposals like facial recognition bans (34%) or universal basic income (25%) fracture the debate.

For now, AI’s promise as a time-saver (55%) and entertainer (51%) outshines faith in its economic impact (36%).

Writing in the report’s foreword, Yolanda Gil and Raymond Perrault, co-directors of the AI Index Report, said: “As AI continues to reshape our lives, the corporate world, and public discourse, the AI Index continues to track its progress—offering an independent, data-driven perspective on AI’s development, adoption, and impact, across time and geography.

“What a year 2024 has been for AI.

“The recognition of AI’s role in advancing humanity’s knowledge is reflected in Nobel prizes in physics and chemistry, and the Turing award for foundational work in reinforcement learning. The once-formidable Turing Test is no longer considered an ambitious goal, having been surpassed by today’s sophisticated systems.

Recommended reading

• PwC Interview | How Agentic AI is Reshaping Business
• Agentic AI on the Rise Despite Workers’ Concerns
• IT Executives Are Primed for Agentic AI

“Meanwhile, AI adoption has accelerated at an unprecedented rate, as millions of people are now using AI on a regular basis both for their professional work and leisure activities. As high-performing, low-cost, and openly available models proliferate, AI’s accessibility and impact are set to expand even further.

“After a brief slowdown, corporate investment in AI rebounded. The number of newly funded generative AI startups nearly tripled, and after years of sluggish uptake, business adoption accelerated significantly in 2024. AI has moved from the margins to become a central driver of business value.

“Governments, too, are ramping up their involvement. Policymakers are no longer just debating AI—they’re investing in it.

“Several countries launched billion-dollar national AI infrastructure initiatives, including major efforts to expand energy capacity to support AI development. Global coordination is increasing, even as local initiatives take shape.

“Yet trust remains a major challenge. Fewer people believe AI companies will safeguard their data, and concerns about fairness and bias persist. Misinformation continues to pose risks, particularly in elections and the proliferation of deepfakes.

“In response, governments are advancing new regulatory frameworks aimed at promoting transparency, accountability, and fairness. Public attitudes are also shifting. While scepticism remains, a global survey in 2024 showed a notable rise in optimism about AI’s potential to deliver broad societal benefits.

“AI is no longer just a story of what’s possible—it’s a story of what’s happening now and how we are collectively shaping the future of humanity. Explore this year’s AI Index report and see for yourself.”