Token Shock是否伤害AI应用？

Mark L. Moerdler, Ph.D.+1 917 344 8506mark.moerdler@bernsteinsg.comRichard Nguyen+33 1 42 13 54 22richard.nguyen@bernsteinsg.comPeter Weed+1 917 344 8390peter.weed@bernsteinsg.comMark Shmulik+1 917 344 8508mark.shmulik@bernsteinsg.comHarshita Rawat, CFA+1 917 344 8485harshita.rawat@bernsteinsg.comStacy A. Rasgon, Ph.D.+1 213 559 5917stacy.rasgon@bernsteinsg.comMark C. Newman+1 212 845 7822mark.newman@bernsteinsg.comNikhil Devnani, CFA+1 917 344 8425nikhil.devnani@bernsteinsg.comMadison Rezaei+1 917 344 8622madison.rezaei@bernsteinsg.comGautam Chhugani+91 226 842 1416gautam.chhugani@bernsteinsg.comLaurent Yoon+1 917 344 8502laurent.yoon@bernsteinsg.comFiroz Valliji, CFA+1 917 344 8316firoz.valliji@bernsteinsg.comShelly Tang, CFA+1 917 344 8342shelly.tang@bernsteinsg.com We don't have all the answers, and frankly we are not sure if wehave all the questions but Bernstein’s US and European Softwareteams have pulled our proverbial heads together and give you ourperspective. Hopefully you find this helpful. By Bernstein’s world-wide software team: Mark L. Moerdler, Richard Nguyen, and Peter Weed GenAI became the buzz across industries starting with the broadpublic launch of ChatGPT v3.5 in late 2022 and v4.0 / v4.0o in H12023 (depending on which model you consider really “first”ito drivebroad awareness / adoption) and since then it has created multipletsunamis shifting global spending toward building AI datacenters;changing how applications are written; concerns that software isgoing to be eaten by AI; creating billionaires and trillionaires and somuch more. All of this investment and disruption has been driven bythe belief that GenAI adoption is going to be huge and while there isno disagreement that adoption is already strong (see Ramp AI Index)the continuing investment cycle argues that Gen AI’s impact will beeven more massive and will be quite impactful across virtually everyindustry and in everyone’s personal lives. WHAT IS THE PROBLEM? HIGH COGS LOW PRICE. The fundamental problem is that GenAI and Agents are not cost free- someone is paying for it. With the launch of ChatGPT followedvery quickly by GenAI capabilities in most existing search tools, allwith generous free to use models, consumers started to becomeaccustomed to free access. It wasn’t just the AI Labs and searchengines… lots of “Born in AI” startups were created with GenAI andAgents built in and, taking a page out of the early days of search, manyof the at least early consumer centric AI-tools were free — lookingto drive “eyeballs” with a plan (or hope) of converting those eyeballsinto revenue at some point in the future. In fact, depending on thetool, there was at least initially no advertisements. Sure, there werepremium editions, but most people could do everything they wantedto without the paid-for premium editions. Even if they were paying, itwas a fixed unlimited usage subscription (until recently) frequently atrecognizable broadly consumer-friendly price levels like $19/29/99per month — the costs didn’t seem excessive for the value delivered.There were discussions of ads, but unless GenAI chatbots wereadded to existing search tools advertising as a monetization strategyis still quite limited. Experts, pundits and industry leaders almost seem to be competingas to how big a number they can create, whether in GPUs sold overthe next few years; datacenter spending; job disrupted; or revenuescreated or lost. Not surprising this has created some roller coastersin the market as one day a company (tech or not) is going to be thebig winner and the next it will be a loser. Even for those with graduate degrees in the space and dozens ofyears in the industry the speed of innovation and the even fasterspeed of commentary is daunting. It wasn’t just consumer looking for free. As GenAI capabilitiesexpandedfrom consumer into commercial the monetizationstrategies of well funded startups and many enterprise app vendors(e.g. Adobe) used free or freemium models. Even if there was a paid In today’s Weekend Tech Byte we are going to focus on a new andpotentially disruptive trend - “Token Shock” driven by the shift of AI toconsumption pricing (tokens) and what it could mean for AI adoption. Some, including some notable tech companies (e.g. Microsoftannounced ~4,800 layoffs or ~2.1% of workforce, Salesforcecut ~4,000 service desk, Meta laid of ~8,000 employees) haveannounced layoffs that included developers (and other roles) but thetotal numbers have been relatively small to date. Not to mention, theyseem to largely impact companies perceived as “over their skis” onhiring, and perhaps “AI washing” necessary and overdue headcountreductions. version or a usage fee for commercial-targeted AI applications itreally was not pushed aggressively initially by either the Cloud or“Born-in-AI” vendor. Companies and their users could largely partakeof the advantages of AI without having to think about the budgetimpact. Again, the plan, we believe, was to drive ado