Double click here to edit new New Header component
Investment Bank
How can we help?
X
  • Contact Us
  • Client Login
    • Search
    Parsys 1
    Double click here to edit Hero Media component
    RESEARCH  |  3 POINT PERSPECTIVE

    The next wave of AI: Demand and adoption

    authorImage authorImage authorImage 
    Contributors: Tom O'Malley, Ross Sandler, Trevor Young
    09 Dec 2024

    Artificial intelligence has quickly captured consumer imagination and enterprise attention. Now, it’s entering a new era, with AI agents set to boost enterprise productivity and assistants to help consumers with everyday tasks. While capital investment requirements will be immense in the coming years, profitability could come sooner than expected, according to our Research analysts.

    Client log-in for full report
    Series Image
    Robot hand
    Parsys 2
    Double click here to edit Go To Section component

    GO TO SECTION

    Parsys 3

    1. AI's next wave is on the horizon

    Since ChatGPT’s debut in November 2022, AI has experienced unprecedented growth: ChatGPT and Meta AI have now gained more than 300 million weekly active users. Other ‘frontier’ or cutting-edge models have also shown promise. 

    Our Research team sees three distinct eras emerging. While chatbots and copilots may still feel new, the text-based Chatbot Era (2023-2024) is about to be superseded: the next wave of AI product innovation is nearly here. 

    It’s important to note that while AI may be making the headlines, broader adoption remains limited, and enterprise use is largely confined to niche applications such as code writing and customer service. The industry has experimented with various business models, but a clear ‘killer app’ has yet to emerge. Most financial gains so far have gone to hardware providers and foundational AI infrastructure. 

    While chatbots and copilots should  continue to grow, the next AI period is poised to begin. The Agent Era (2025-2026) is likely to be defined by AI systems that can complete end-to-end tasks autonomously, in contrast to the previous chatbot and copilot stage of AI. These agents are likely to bring about meaningful consumer and enterprise AI adoption with far more accurate and powerful responses. Agents are expected to boost enterprise productivity and help consumers with everyday tasks such as completing transactions.  

    As the quality and performance of AI applications increases by embedding agents, our Research team predicts that consumer AI adoption should hit one billion daily active users (DAU) by 2026. However, significant challenges remain if this target is to be achieved, including reducing errors and improving memory. There are some concerns that the concept of ‘scaling laws’, which posits that AI models improve with more compute and data, may be plateauing, potentially signaling leveling out in the growth and pace of AI advancement. However, AI insiders dispute this perspective.

    Looking further ahead, the AI Employee and Consumer Robot Era (2027-2028) envisions AI evolving into digital employees for enterprises and humanoid robots for consumers. These advancements could reshape industries by reducing human involvement in tasks and increasing efficiency. Digital employees and humanoid robots are still in early development and likely years away from widespread adoption.

    2. And adoption is set to soar 

    ChatGPT and other consumer text-based tools have achieved significant user adoption in a short time, and AI is starting to proliferate in industries and sectors where a clear return on investment exists, such as marketing, analytics and engineering

    As we move on to the next AI eras, Barclays Research expects huge growth. While consumer AI use should continue to grow rapidly, enterprise is where the action is likely to be. Our analysts estimate that there are around 7 billion tasks that AI agents could execute independently, enabling entire teams and organisations to increase system-wide productivity. 

    While it will take time for AI systems to gain ground, the pace of adoption is likely to accelerate rapidly: starting with just 2.5% of enterprise tasks in 2025, AI could capture 5% of this market by 2026 and could add an additional 250 to 500 basis points of task adoption annually.

    As AI adoption grows, so should the demand for computational resources such as graphics processing units (GPUs) and application-specific integrated circuits (ASICs). These are essential for training models and putting them into operation to make predictions or draw conclusions based on unfamiliar data (a process known as inference).

    Our Research team expects demand for GPUs and ASICs will be 250% higher than consensus forecasts in 2025 and rise to 14 times current levels by 2027. This surge will be partly driven by the proliferation of AI agents, which require significant compute power, as each human query is autonomously passed through an AI platform multiple times. For example, even a modest adoption of AI agents for 2% of tasks in 2025 could increase inference compute needs by 15%. 

    3. Costs are falling fast 

    While demand is increasing rapidly, costs are falling fast. Inference unit costs per 1 million tokens (the smallest unit of data that is processed by algorithms) have fallen by 90% in the past 18 months and are expected to continue dropping.  

    Lower AI costs are driven by factors such as algorithmic improvements, competition and a shift from GPUs to custom ASICs for inference workloads. Major AI players like Amazon, Meta, Google and Microsoft, known as hyperscalers, are developing expensive custom ASICs for training and AI inference which could outperform GPUs for specific tasks on price/performance.

    Barclays Research estimates that GPUs currently handle 80% of the inference market, but their share is expected to drop to around 50% by 2028 as hyperscalers’ custom ASICs gain market share.

     

    The capital investment required for AI remains immense. Chip-related capital expenditure for consumer AI inference alone is expected to approach $120 billion in 2026 and exceed $1.1 trillion by 2028.

    However, falling costs are expected to significantly improve AI's unit economics. According to our Research team’s analysis, some leading AI companies are already profitable on a per-model basis (excluding future model development costs). As model development slows in the next 5–10 years and inference costs decrease, profits could grow substantially. This suggests current concerns about losses and high burn rates may be overstated.

     

    Important content disclosures

    Parsys 4
    Double click here to edit Banner component
    Double click here to edit Experts Panel Card component

    About the experts

    Ross Sandler

    Managing Director, Senior Equity Research Analyst

    Ross Sandler is a Managing Director and Senior Equity Research analyst for the Internet sector, which he has covered since 2004. Institutional Investor has ranked him among the top 10 Internet analysts since 2011. His team is responsible for coverage of the global consumer internet landscape including e-commerce, digital advertising, mobile, online travel, and video games, including companies listed in the US and China. Based in San Francisco, Mr. Sandler joined Barclays in 2017 after four years as a managing director in Internet and Digital Media at Deutsche Bank. Prior to Deutsche Bank, he spent eight years with RBC and four years at UBS. He earned a BS in Business Administration from the University of New Hampshire and an MBA from New York University.

    Tom O'Malley

    Director, Semiconductor Research Analyst

    Tom O'Malley is a Director and Equity Research analyst covering U.S. Semiconductors & Semiconductor Capital Equipment. He joined Barclays in 2014 as a member of the Semiconductor team and went on to lead coverage of SMID Cap Semis. Tom took over coverage of the Semiconductor & Semiconductor Capital Equipment franchise at the end of 2023. Prior to joining Barclays, Tom worked at Ernst & Young in Advisory Services and earned a B.S. in Finance and English minor from Lehigh University.

    Trevor Young, CFA

    Director, Internet Research Analyst

    Trevor Young is a Director and Equity Research analyst covering U.S. Internet. He is responsible for Barclays' U.S. SMID-cap internet coverage, including online travel, e-commerce, web tools, and SMB services. Trevor has been part of top 10 ranked teams in the Institutional Investor poll throughout his career. Prior to joining Barclays in 2017, Trevor worked at Jefferies covering consumer discretionary and leisure names. Trevor spent nearly a decade in real estate private equity before moving into equity research. He is a CFA Charterholder and a former professional mountain bike rider.

    Parsys 5
    iParsys for Double Pixel component