10 Charts That Define the AI Boom, According to Mary Meeker

IN THIS ISSUE: Dig into key insights from Mary Meeker’s latest AI report, a sweeping 340-slide analysis that breaks down where adoption is accelerating, how innovation is evolving, and what it means for the future of technology.

The Prompt

Investor Mary Meeker’s reports are among the most closely watched in the industry, with entrepreneurs, investors, businesses, and the media poring over every chart and insight. The Bond VC general partner’s Internet Trends report has been frequently cited and provides a clear snapshot of tech’s future. And while it’s been six years since her last trends report, Meeker’s firm released one focused exclusively on artificial intelligence—and it’s a whopper of a read.

Coming in at 340 pages, the Trends in Artificial Intelligence report explores AI’s foundational trends, covering the “unprecedented” growth, user and developer adoption, rising competition, the battle with China, and more. “We’ve never seen anything like the user growth of ChatGPT, particularly outside the U.S., and it shows how the global dynamics of tech and distribution have changed,” Meeker tells Axios. She believes AI is growing at a faster pace than other technological shifts, such as the PC, desktop internet, mobile internet, and cloud computing.

Given the breakneck pace of AI development, even the most up-to-date insights can become outdated within months. Some of the data Meeker cites is from 2024 or early 2025, which means parts of this report may already feel dated by the time you’re reading it. That’s not necessarily a flaw—it’s a testament to just how rapidly the landscape is evolving. In the world of AI, blink and everything could look different.

Here are ten slides that capture some of the most important takeaways from the report:

1. AI Spurring CapEx Spending by Big Tech

Artificial intelligence is forcing Big Tech firms (i.e., Apple, Nvidia, Microsoft, Alphabet, Amazon, and Meta) to invest heavily in capital expenditures. In the decade since 2014, spending among these companies has increased by more than 21 percent annually, driven by the dramatic rise in global data generation, a trend that has grown at a rate of more than 28 percent per year. And as these firms gobble up more data, they’ll need to put more dollars into building more hyperscale data centers, faster network infrastructure, and significantly more compute capacity to process and leverage this immense influx of data. 

Why it matters: This chart highlights the fundamental investment required to drive the AI revolution. The massive amount of capital deployed isn’t just about expansion; it’s about accommodating the surge in AI demand, which thrives on vast datasets and immense computational power. 

As Meeker notes, “The world’s biggest tech companies are spending tens of billions annually – not just to gather data, but to learn from it, reason with it and monetize it in real time. It’s still about data – but now, the advantage goes to those who can train on it fastest, personalize it deepest, and deploy it widest.

2. AI Models Devour Data as Set Sizes Skyrocket

Over the past 15 years, the size of datasets used to train AI models has ballooned by more than 250 percent annually, according to Epoch AI. This graph highlights a consistent and rapid increase in the number of tokens used for training. Notably, many of the post-2022 models tend to have massive training datasets. This isn’t a coincidence; it’s the result of multiple factors, including the significant capex investments being made, the emergence of foundational models and LLMs with emergent capabilities not typically seen in smaller models, and the sheer volume of data and careful curation.

Why it matters: Size and diversity of training data are crucial determinants of an AI model’s capabilities and performance. The 250 percent annual growth highlights the growing demand for data in AI development. Having access to a vast amount of information has become a critical competitive advantage. Meeker’s chart suggests that as new models emerge, their datasets will continue to be enormous, furthering advancements and applications, but could raise questions about data collection, curation, and computational resources needed to handle such immense scales.

3. The Energy Footprint of AI

It’s known that we have an energy crisis caused by the surge in AI usage. Companies like Microsoft, xAI, and Meta are acquiring nuclear power or building dedicated factories to supply the energy needed for their data centers. In fact, hyperscalers are the primary drivers of the overall surge in energy consumption over the past few years. More specifically, it’s the servers that need the power. 

Why it matters: While the explosive growth of AI is unlocking unprecedented capabilities, it comes with a steep environmental cost. The massive and rising energy demands of data centers powering AI and hyperscale operations pose serious challenges to sustainability and grid reliability. Meeker’s report highlights these concerns as part of the broader AI trend narrative. As AI adoption accelerates across industries, it’s critical to address its energy footprint, ensuring that advances in intelligence can be met by renewable power and a resilient energy infrastructure. 

4. Will We Achieve a ZIRP for Inference?

It’s becoming significantly cheaper to run inference in AI models. Meeker finds that the cost of making AI perform a task has plummeted by 99.7 percent over the past two years. When OpenAI’s GPT-3.5 debuted in late 2022, it had an inference price of more than $10 per million tokens. Its successor, GPT-4-0314 had a price tag of nearly $20 per million tokens. Fast forward to May 2024, and inference for GPT-4o-2024-05 would cost under $10 per million tokens. DeepSeek-V3, which launched at the end of 2024, costs under $1 per million tokens. 

Why it matters: AI innovations have led to rapid commoditization of model serving, making capabilities more affordable and accessible. As a result, it’s lowering a key barrier to entry for developers looking to use AI in their products and services. With these cost savings, it is hoped that they’ll lead to more innovation, greater AI adoption, and increased access to powerful AI tools.

5. AI Companies Can Monetize Faster Than SaaS Companies

When comparing 100 AI companies versus Software-as-a-Service (SaaS) companies, the former was able to reach $5 million in annualized revenue at a much faster rate than the latter, 24 months versus 37 months. 

Why it matters: AI is disrupting the commercialization landscape. The accelerated revenue growth for AI companies suggests there’s a strong market appetite for these solutions, perhaps because of the immediate value and efficiency gains they provide. 

6. Closed vs. Open-Source Models

Which is better: Open-source AI or closed? That’s an ongoing debate in technology, but according to Meeker’s report, when it comes to where more computing resources are being devoted, closed models have the advantage. There’s a 17-month lag between closed and open models, meaning that the latter typically could achieve a similar level of compute intensity about 17 months after their closed-source counterparts. Moreover, closed models are said to have a steeper growth rate in compute (5.2x per year) compared to their open-source peers (3.6x per year).

Why it matters: Hugging Face, Ai2, and others in the open-source community have been advocating for companies to share their AI software. However, this trend suggests that companies capable of higher computer investment will have access to the most cutting-edge and powerful models, which will enable more sophisticated applications and services. As a result, AI superiority could be relegated to a handful of players instead of a larger community. The extreme cost could also be a prohibitive barrier to entry for smaller businesses and researchers. The “17-month lag” isn’t something to avoid either. In a way, it’s giving closed models a distinct advantage over open-source competitors. 

7. U.S. and China Dominate AI

How close is China to the United States when it comes to artificial intelligence? Both countries are leaders in large-scale AI systems, outperforming the United Kingdom, France, Canada, Germany, and others. America ranks number one with nearly 150 such systems, but China is catching up with just over 100.

Why it matters: This slide highlights that the U.S. and China have established themselves as clear leaders in AI. It has profound implications for global economic competitiveness, national security, and technological sovereignty. For the rest of the world, they’ll likely be reliant on either American or Chinese AI technologies, raising concerns about data privacy, ethical alignment, and geopolitical leverage. This competition is why the U.S. has opted to support efforts in Saudi Arabia to establish data centers in the kingdom and has banned the sale of AI chips to Chinese companies.

8. Now Hiring: AI

Having AI skills is becoming a prerequisite for securing a job in today’s market. According to the University of Maryland and LinkUp, the number of AI job postings in the U.S. has jumped by more than 448 percent over the past seven years, while non-AI IT jobs have dropped by nine percent during that same period. 

Why it matters: This underscores a fundamental transformation in the labor market, driven by the increasing adoption and advancement of AI. The explosive growth in AI job postings reflects a surging demand for specialized skills related to AI development, deployment, and maintenance. For businesses, this underscores the urgent need to invest in AI talent and upskill their existing workforce. Staying competitive in this technological era will require not only hiring AI-savvy professionals but also providing current employees with access to reskilling programs such as Salesforce’s Trailhead, ServiceNow University, LinkedIn Learning, and other L&D platforms.

9. It’s Easier for AI to Be Mistaken for Human

As AI models get more sophisticated and are imbued with persona capabilities, it’s becoming more challenging to identify whether the content is from a human or a machine. In April, OpenAI’s GPT-4.5 model passed the Turing Test, a barometer for human-like intelligence. This means that testers believed the LLM’s responses were human 73 percent of the time when it adopted a persona. This far surpassed other leading models, including GPT-4o and Meta’s Llama 3.1-405B.

Why it matters: It’s a double-edged sword. Having AI sound and act more human is good for businesses and individuals who want the tech to act on their behalf. Having a model pass the Turing test demonstrates AI’s ability to engage in complex, human-like conversation. However, it also raises concerns about ethics and trust. 

10. ChatGPT Shows Unprecedented AI Global Adoption

Meeker doesn’t hold back in calling AI’s rise “unprecedented”—a word she uses no fewer than 51 times throughout the report. But she’s not exaggerating. OpenAI’s ChatGPT, the poster child of this AI era, serves as her prime example of just how quickly the technology has been embraced. It’s the fastest app to top 100 million users in the past two decades. It only took five days for it to reach one million users, compared to the iPhone (74 days), and at a substantially lower price point ($0). Additionally, as for its global adoption, ChatGPT achieved 90 percent adoption in just three years. By comparison, the internet took 23 years to reach that feat and back then, it was considered to be one of the fastest technological rollouts in history.

Why it matters: The speed of AI adoption is amazing, and this proves it isn’t a niche technology, but rather a viral phenomenon that’s becoming a part of our everyday digital lives. The widespread adoption worldwide indicates that AI will play a significant role in our economies, societal norms, and regulatory frameworks.

▶ You can read Mary Meeker’s full report, Trends in Artificial Intelligence, here.

---

Don’t Miss out on Future Issues of ‘The AI Economy’

The AI Economy is expanding! While you’ve been getting weekly insights on LinkedIn, I’m gearing up to bring you even more—deep dives into AI breakthroughs, more interviews with industry leaders and entrepreneurs, and in-depth looks at the startups shaping the future. To ensure you don’t miss a thing, subscribe now on Substack, where we’ll be rolling out more frequent updates.

Don’t worry; the weekly newsletter will still be published on LinkedIn, but other stories will be available on Substack.

Stay ahead in the AI revolution—sign up today!

---

This Week’s AI News

🏭 AI Trends and Industry Impact

• Company boards push CEOs to replace IT workers with AI (CIO)
• Google DeepMind’s CEO thinks AI will make humans less selfish (Wired)
• China is gaining ground in the global race to develop AI agents (Rest of World)
• DreamWorks co-founder Jeffrey Katzenberg likens AI to CGI revolution (Axios)
• A dystopian “documentary” about the incomplete repetitive lives of AI actors in between prompts (Laughing Squid)

🤖 AI Models and Technologies

• Ai2’s RewardBench 2 is a tougher benchmark for testing how well AI models reflect human judgment (My Two Cents)
• Chatbot platform Character.ai unveils video generation, social feeds (TechCrunch)
• DeepSeek reportedly may have used Google’s Gemini to train its latest model (TechCrunch)

✏️ Generative AI and Content Creation

• OpenAI’s ChatGPT can now read your Google Drive and Dropbox (The Verge)
• Mistral releases a vibe coding client, Mistral Code (TechCrunch)
• Microsoft integrates OpenAI’s Sora video creator into Bing (Engadget)

💰 Funding and Investments

• North America takes the bulk of AI VC investment, despite tough political environment (TechCrunch)
• Anysphere, the company behind the vibe coding agent Cursor, raises $900 million at a $9.9 billion valuation (Bloomberg)
• Cohere reportedly looking to raise $500 million as AI race speeds up (Crunchbase)
• Mind raises $30 million to help businesses prevent data leaks using AI (Silicon Angle)

☁️ Enterprise AI Solutions

• OpenAI hits 3 million business users and launches workplace tools to take on Microsoft (VentureBeat)
• Snowflake’s Openflow tackles AI’s toughest engineering challenge: Data ingestion at scale (VentureBeat)
• Snowflake’s new AI agents make it easier for businesses to make sense of their data (ZDNET)
• Augment Code’s remote agents code in the cloud (The New Stack)
• HubSpot connects its CRM with ChatGPT to deliver smarter customer insights (My Two Cents)
• Adobe launches new product support AI agent to lighten the customer load (CX Today)
• Amazon is in talks to roll out AI coding assistant Cursor internally as employee interest spikes (Business Insider)
• Aptori is building an agentic AI security engineer (The New Stack)

⚙️ Hardware, Robotics, and Autonomous Systems

• Amazon is reportedly developing software for humanoid robots that will deliver packages (The Information)
• AMD takes aim at Nvidia’s AI hardware dominance with Brium acquisition (TechCrunch)
• AMD acquires the team behind AI chip startup Untether AI (CRN)

🔬 Science and Breakthroughs

• FDA launches agencywide AI tool to help scientific reviewers, investigators, and others streamline their work (Axios)

💼 Business, Marketing, Media, and Consumer Applications

• Meta aims to fully automate ad creation using AI (The Wall Street Journal)
• AI-driven search ad spending set to surge to $26 billion by 2029: Study (Reuters)
• Amazon’s Lab126 hardware R&D lab forms new agentic AI group (CNBC)
• Everyone in Hollywood is already using AI (and hiding it) (Vulture)
• AMC Networks inks deal with generative AI firm Runway (The Hollywood Reporter)
• How Sam’s Club CTO used computer vision and AI to transform the member exit experience (Fast Forward)
• Salesforce acquires Moonhub team as AI hiring arms race escalates (My Two Cents)

⚖️ Legal, Regulatory, and Ethical Issues

• Reddit sues Anthropic for breach of contract and “unfair competition” (CNBC)
• Machine learning pioneer Yoshua Bengio launches $30 million nonprofit to rethink AI safety (Axios)

💥 Disruption, Misinformation, and Risks

• The rise of “vibe hacking” is the next AI nightmare (Wired)
• No AI, no job. These companies are requiring workers to use the tech (The Washington Post)
• The hottest new vibe coding startup may be a sitting duck for hackers (Semafor)
• McKinsey is using AI to create PowerPoints and take over junior employee tasks: “Technology could do that” (Entrepreneur)

🔎 Opinions, Analysis, and Editorials

• Anthropic CEO Dario Amodei: Don’t let AI companies off the hook (The New York Times)
• How much information do LLMs really memorize? Now we know. (VentureBeat)
• I’m Expedia’s marketing chief. Here’s how we’re preparing for a future when people use AI to plan their vacations (Business Insider)
• AI killed the math brain (The New York Times)
• Why AI won’t kill programming (TechTalks)

---

End Output

Thanks for reading. Be sure to subscribe so you don’t miss any future issues of this newsletter.

Did you miss any AI articles this week? Fret not; I’m curating the big stories in my Flipboard Magazine, “The AI Economy.”

Connect with me on LinkedIn and check out my blog to read more insights and thoughts on business and technology. 

Do you have a story you think would be a great fit for “The AI Economy”? Awesome! Shoot me a message – I’m all ears!

Until next time, stay curious!