Global Internet: Commercialization Framework of AI Models and Market Segmentation

Global Internet: A framework for AI model commoditisation, and A shift in the AI zeitgeist?We’re long-term AI bulls, and have increasingly built agenticAI into our day-to-day workflows. That said, several recent events strike us as beingnoteworthy as far as AI competition and ROI is concerned. The high cost of Anthropic’sClaude Fable 5 pricing causing devs to ration usage feels like a notable development. Seealso headlines pointing to OpenAI potentially cutting prices to win share, and large AI users Robin Zhu+852 2123 2659robin.zhu@bernsteinsg.com Mark Shmulik+1 917 344 8508mark.shmulik@bernsteinsg.com An alternative mental model for AI model commoditisation.Rather than modelintelligence necessarily converging, we increasingly think AI model commoditisation will be driven byhuman user perceptionacross various end uses, and the timeline on whichcompeting models become “good enough”, and reliable at scale. We’ve long argued thatconsumer-focused AI (ordering bubble tea, booking hotel rooms) will commoditise relativelyquickly. Enterprise use cases like coding and Excel work are more complex, and will take Charles Gou+852 2123 2618charles.gou@bernsteinsg.com Min-Joo Kang+852 2123 2644minjoo.kang@bernsteinsg.com Hyrum Caesar+81 3 6777 6979hyrum.caesar@bernsteinsg.com The implications for AI market segmentation… and competition.Once theAI supporting a given use case becomes “good enough”, we’d expect the arbiter of competition and market share to shift from reasoning capabilities to availability and price.Our base case is the US frontier labs continue to unlock new model capabilities thatcommand a rising price premium among increasingly specialised customers. Data securityconcerns and global geopolitics likely prevent some US and European enterprises from Wenhuan Chang+1 917 344 8546wenhuan.chang@bernsteinsg.com Deeksha Pandey+1 917 344 8447deeksha.pandey@bernsteinsg.com Chinese open source models as the value option.Hardware advancements will makeit cheaper for the US frontier labs to serve tokens, and drive Jevon’s Paradox. Nonetheless,it would surprise us if trailing edge AIdidn'tbecome another sector where Chinesecompetition takes market share, in part thanks to a willingness to accept lower prices andmargins. The ability to follow global SOTA as a lighthouse for R&D direction means theChinese AI labs need to do less exploratory research. Our market segmentation suggests Stepping off the exponential - and the path to AI lab profits.If we’re right about AIcommoditisation being driven by task completion and user perception, the implications forAI lab economics might actually be reasonably hopeful. Once a use case becomes “solved”the return on doing incremental R&D on the same topic diminishes, meaning R&D effortcan be redirected elsewhere along the frontier. Over time, we’d expect the envelope of BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We’re AI bulls at a high level, but think events in the past week are worthy of attention. The high cost of Claude Fable 5 tokensmaking token-maxxing uneconomic wasn’t necessarily on our list of high-likelihood events for the year, but likely acceleratesthe timeline on which AI devs and users start to examine AI ROI more closely. The idea that AI users will select across a range ofmodels to match the marginal cost of tokens to the marginal benefit of task completion (with considerations perhaps given to To take things one step further, our mental framework for AI model commoditisation focuses on whether task completionbecomes reliable at scaleas perceived by human users, as opposed to some academic measure of underlying convergence. Once agentic AI “solves” various vertical use cases, we’d expect the arbiter of competition to shift from reasoning capabilitiesand task completion to more mundane variables like cost per task, reliability, and developer trust. The debate on whetherthe top Chinese AI models are 6 or 12 months behind US SOTA is intellectually interesting, but probably almost meaningless Longer term, we think it’s logical to assume a two-tier market split across (1) the US SOTA labs serving the frontier, where theworkloads become increasingly specialised and novel, while a narrower field of customers exhibits high willingness to pay; and(2) a market for trailing edge AI that serves more mundane enterprise and consumer end uses, where the availability and cost ofreasoning drives adoption - and share shifts. Consumer-oriented AI should be one of the first end uses where a wide range of VALUATION COMPS TABLE DETAILS A MENTAL MODEL FOR AI MODEL SEGMENTATION AND COMMODITISATION The debate over whether AI models can remain differentiated or are doomed to commoditise in the long run remains a key onewhen it comes to the economic value associated with frontier intelligence, and the valuations of market bellwethers like OpenAIand Anthropic. Our own views on this topic have meandered over time. But we increasingly think a generalisable framework for