If the adoption of AI feels totally different from any tech revolution you might have skilled earlier than — cellular, social, cloud computing — it really is.
Enterprise capitalist Mary Meeker simply dropped a 340-page slideshow report — which used the phrase “unprecedented” on 51 of these pages — to explain the velocity at which AI is being developed, adopted, spent on, and used, backed up with chart after chart.
“The tempo and scope of change associated to the factitious intelligence expertise evolution is certainly unprecedented, as supported by the info,” she writes within the report, referred to as “Developments — Synthetic Intelligence.”
There’s a sure poetic historical past to this particular person writing this sort of report. Meeker is the founder and common associate at VC agency Bond and was as soon as often called Queen of the Web for her earlier annual Web Developments experiences. Earlier than founding Bond, she ran Kleiner Perkins’ development apply, from 2010-2019, the place she backed corporations like Fb, Spotify, Ring, and Block (then Sq.).
She hasn’t launched a tendencies report since 2019. However she dusted off her abilities to doc, in laser element, how AI adoption has outpaced another tech in human historical past.
ChatGPT reaching 800 million customers in 17 months: unprecedented. The variety of corporations and the speed at which so many others are hitting excessive annual recurring income charges: additionally unprecedented.
The velocity at which prices of utilization are dropping: unprecedented. Whereas the prices of coaching a mannequin (additionally unprecedented) is as much as $1 billion, inference prices — for instance, these paying to make use of the tech — has already dropped 99% over two years, when calculating price per 1 million tokens, she writes, citing analysis from Stanford.
The tempo at which rivals are matching one another’s options, at a fraction of the associated fee, together with open supply choices, significantly Chinese language fashions: unprecedented. For instance, she factors out that Nvidia’s 2024 Blackwell GPU makes use of 105,000x much less vitality per token than the corporate’s 2014 Kepler GPU predecessor.
In the meantime, chips from Google, like its TPU (tensor processing unit), and Amazon’s Trainium, are being developed at scale for his or her clouds — that’s shifting shortly, too. “These aren’t facet tasks — they’re foundational bets,” she writes.
The one space the place AI hasn’t outpaced each different tech revolution is in monetary returns. Whereas VCs are pouring cash on the AI fireplace as quick as they’ll, AI corporations and cloud service suppliers are additionally burning by way of money. AI requires large investments in infrastructure.
That’s good for shoppers and enterprises, the beneficiaries of quick enhancements, whereas competitors lowers prices, Meeker factors out. However the jury remains to be out over which of the present crop of corporations will turn out to be long-term, worthwhile, next-generation tech giants. “Solely time will inform which facet of the money-making equation the present AI aspirants will land,” she writes.
As for the remainder of us: Simply maintain on to your hats.
