Mary Meeker, a high-profile analyst known as "the Queen of the Internet" because of her early, bullish calls on the prospects for Amazon, Google and Apple and then for the massive yearly reports she issued on the state of the internet, has turned her attention to AI.
In her first major report in five years, Meeker makes the case that AI has been improving far faster even than we realize and that progress is accelerating, with no end in sight.
Her forecast — and she's generally right — should be very good news for insurers.
Meeker's report, all 339 data-and-graphics-packed pages of it, uses the word "unprecedented" dozens and dozens of times. I'll spare you the detail, but here are a few nuggets you might want to include in any presentation arguing for investment in AI:
- ChatGPT hit one billion searches per day in less than two years. Google needed 11 years to reach that mark.
- While developing and training generative AI models is wildly expensive, the cost of using AI has declined 99% just in the past two years.
- The “Big Six” U.S. tech giants (Apple, NVIDIA, Microsoft, Alphabet, Amazon and Meta) are going to keep spending unfathomable sums to improve capabilities and drive costs down. They spent $212 billion on capital expenditures in 2024, up 63% from 2023, largely on AI chips, data centers, and cloud infrastructure. They are investing 13% of revenue on R&D, up from 9% a decade ago, and they have the resources to keep going: Their annual free cash flow is nearly $400 billion.
- While there is a lot of concern about the energy consumption of AI, Nvidia's latest 2024 Blackwell GPU achieves 105,000 times greater energy-efficiency than its 2014 predecessor.
Meeker also offers many useful examples of how generative AI is, and will, find its way into the real world:
- More than 10,000 doctors at Kaiser Permanente use an AI assistant to automatically document patient visits, freeing three hours a week for 25,000 clinicians.
- Stripe pushed one important fraud-pattern catch rate from 57% to 97% overnight.
- 27% of ride-hailing trips in San Francisco are handled by autonomous vehicles. (That's much higher than I would have guessed.)
- By early 2025,, evaluators thought 73% of the output from a GPT was written by humans.
Meeker's projections for five and 10 years out may be even more startling. By 2030, she expects that AI will:
- Generate human-level text, code and logic, in any number of languages.
- Run autonomous customer service and sales.
- Collaborate like a creative partner.
By 2035, she says AI will:
- Conduct scientific research
- Design advanced technologies
- Simulate human-like minds
- Operate autonomous companies
- Perform complex physical tasks in real-world environments
Any projections for technology that reach out a decade often verge on science fiction or at least fuzzy optimism -- a lot of projections about AI, for instance, were "coming in 10 years" for decades. But Meeker paints a picture of intriguing possibilities that we should all explore.
A lengthy analysis of her report on Substack offers an even rosier outlook for insurers. It says that, while many companies and jobs will be overtaken by the growing power of AI, it won't threaten businesses that have these three levers:
- "Data gravity – proprietary or regulated corpuses (medical imaging, trade documents, tele-metrics) that outsiders cannot legally pull into pre-training.
- "Reward ambiguity – industries where you can underwrite the outcome (fraud risk, quality-of-care scores, turbine uptime) and price on financial exposure.... Risk pays!
- "Compliance bottlenecks – any workflow where passing the audit is the moat."
That sure sounds like insurance to me: proprietary data, underwriting of risk, and compliance bottlenecks. So insurers can take advantage of the huge amounts of horsepower that the gen AI model companies are providing, while secure in the knowledge of the health of the underlying business.
Insurers can now start to raise their sights. At the Instech conference in New York City last week, where I had the pleasure of speaking, I heard about remarkable improvements in data intake from Concirrus, Cytora, and Federato, based on AI engines. By next year, I'd expect to hear about similar progress in the assistants that companies are building for underwriters, claims representatives, and agents and brokers so they can process more information faster and make better, more consistent decisions. The year after that, I'll bet we're hearing about whole streams of work being automated through AI.
In time, I suspect we'll stop even talking about chatbots because the capabilities will be built into everything, making AI essentially the user interface for companies. We'll just go to a website or make a call and ask a question. AI will then provide a summary answer, much as Google and other search engines are now doing, and offer next steps.
Eventually, the arms race by gen AI companies may slow, as losers drop out of the competition. At that point, prices for us users could rise, or at least stop plummeting, following much the same dynamic that saw Uber and Lyft raise prices after years of subsidizing rides to lock in interest among riders and drivers.
But Meeker makes clear that any slowing won't come any time soon. For the next few years, at least, it's full speed ahead.
Cheers,
Paul