M FoundationThematics | EuropeAI: Adopters Are AdoptingMorgan Stanley & Co. International plc+Edward StanleyEquity Strategist Edward.Stanley@morganstanley.com +44 20 7425-0840 Matias OvrumEquity Strategist Matias.Ovrum@morganstanley.com +44 20 7425-9902 Exhibit 2 : Rebased forward P/E of Adopters and Enabler/Adopters re-rating YTD98100102104106108110Nov 23Nov 23Nov 23Nov 23Nov 23Dec 23Dec 23Dec 23Dec 23Jan 24Jan 24Jan 24Jan 24Feb 24Enabler / Adopter BothAdopter Source: FactSet, Morgan Stanley ResearchWe have argued that 2024 will be the "Year of the Adopters". The market is still treating this thesis as a "show me" story; Enablers are still driving returns. Yet, be it Consumer or - more importantly - Enterprise adoption, monthly data points we track suggest AI adoption is continuing at pace. "Once an innovation delivers a competitive advantage everyone must either adopt it, leapfrog it, switch focus or lose market share and eventually go bust." Mustafa Suleyman, Co-Founder of DeepMind. This comment neatly encapsulates the thinking and urgency behind our call for 2024 being the "Year of the Adopter". No one individual datapoint can answer the question of where we are in this journey, but by joining various consumer and enterprise dots, we are able to gauge direction of travel and momentum:• First movers are causing divergences: There are a widening array of consumer-facing chatbots (ChatGPT, Bard/Gemini, Claude, Perplexity etc) and yet OpenAI's lead with consumers remains commanding...albeit yet to make new highs after the Christmas break ( Exhibit 7 ). Developers present the best case study for early adoption. The prospect of >50% productivity gains is driving ever increasing traffic to Github Copilot even as Stack Overflow responded by integrating its own AI assistant in 4Q23 ( Exhibit 1 ). • Open source model listings surpass half a million: From a more enterprise perspective, models listed on Hugging Face - an open source model repository - have surpassed 500,000 and datasets used for fine-tuning them have surpassed 100,000. The people or companies downloading these models are likely using them as starting points for their own domain-specific models, fine-tuning of which requires GPU capacity ( Exhibit 1 ). • Open source model downloads by Adopters rising 120% YoY: While anyone can download these models we consider the direction of travel as somewhat of a proxy for enterprise trial-and-error and/or Adoption. We cannot deduce when prototype moves to production. Nonetheless, model downloads are running at 120% YoY, but of which the top 10 models are seeing the lions' share of downloads (>53% share) and running at a 300% YoY download growth rate ( Exhibit 16 ). • Text models now less than 50% of models vs multi-modal: Multi-modal models (i.e. those capable of ingesting and outputting a combination of text, image, audio and/or video)...
