- There is a school of thought that AI is overhyped and headed into the third ‘AI winter‘ (the first two were 1974-80 and 1987-2000).
- Despite the shortcomings of current LLMs (eg Reversal Curse), we believe there is enough momentum for a long adoption runway.
- In this note we dive into key AI topics, surveying the technologies and their implications for investors and companies alike. We identify a number of UK companies well placed to benefit.
The five themes explored within revolve around adapting AI, budgets, environmental impact, edge AI being a better bet, and what prevents enterprises from taking full advantage of their data. This thorough survey aims to replicate the predictive success of our Generation 3.0 note.
AI is here, and is turbo-charging productivity. “Our biggest concern is not understanding the full implication of GenAI for our business, which has been very slow to respond to technological changes” – CRO of an energy company. We think a lot of investors and companies are too complacent, and fail to recognise the opportunities to take advantage of, and pitfalls to avoid. Even the regulatory changes (eg the EU AI Act) get superficial coverage. The current wave of AI innovations goes beyond the predictive AI of the 2010s, unlocking creative output (eg writing, design, coding), enabling better functionality (eg contextual search and summarisation), upgrading experiences (eg human-like chatbots), and turbo-charging decision-making (eg smart assistants).
In its most recent results, Microsoft noted that its Copilot, targeting programmers at $30/month, had seen 40% QoQ growth to 1m paid users across 37k organisations. We expect similar take-up across a broader range of use cases over the coming years, changing competitive dynamics across many industries. There is a long list of AI beneficiaries based out of the UK. Aligned to our five themes, we note that UK-based companies such as Raspberry Pi, Cambridge Mechatronics and XMOS in edge AI, Matillion and FD Tech in the data stack, and Softcat and Bytes as far as overall AI budgets are concerned, are poised to benefit as the market shifts towards them.
Executive summary
UK investors, across both private and public markets, need to remain well versed in the rapidly changing AI landscape. We dive into five key AI themes to understand how AI models are evolving, what companies are doing with AI-related IT budgets, question AI’s environmental impact, recognise there is a better hardware opportunity beyond NVIDIA, and explain why the data stack is not quite ready for AI to take advantage of enterprise data.
UK is home to many disruptive AI-focused startups: Wayve – next generation self-driving, Synthesia – generative AI clones, Gensyn – decentralised AI compute to dethrone hyper-scalers, Seldon – Machine Learning ops platform, etc. Even excluding the long list of AI-powered healthtech and fintech companies, UK’s AI startup list is long. This is partly down to great universities, which for example have seen a majority of spinout grants go to AI-related activities.
However, one of the criticisms we face in the UK is the perceived lack of deep institutional understanding of relevant AI topics. While we do not buy into this view, we have for years educated the market on generational technologies. Our February 2018 Generation 3.0 note, which foretold the deep learning revolution, is one such example. In surveying the current AI landscape, we have decided to focus on topics that broadly fall within the following five themes.
1. Adaptation of general-purpose AI to specific domains:
a. Companies transform general-purpose large language models (LLMs) using proprietary datasets.
b. These specialized LLMs outperform general ones (like OpenAI, Anthropic) for enterprise-specific workloads.
c. Companies leverage domain-specific LLMs in fields like customer service and video game services.
d. Techniques such as retrieval-augmented generation (RAG) help onboard Enterprises lacking resources to fine-tune LLMs.
e. Emerging markets include software and services that enable domain specificity, like vector databases (KDB.ai) and private LLM hosting (eg Kao Data).
2. Increasing tangibility of AI budgets:
a. Decision-makers are initially hesitant about LLMs due to trust and resource issues.
b. Evidence of productivity boosts from GenAI are starting to change this attitude.
c. Tools like GitHub Copilot show growing usage, due to a significant productivity boost with a clear ROI.
d. GenAI acts as a performance leveller, aiding lower performers significantly more than top performers.
e. Resellers like Bytes and Softcat are key players set to benefit from these expanding AI budgets.
3. AI’s impact on the environment is overstated:
a. Contrary to concerns, AI’s rise does not necessarily lead to higher energy demands and CO2 emissions.
b. Efficiency gains in hardware and software, and hyper-scalers’ focus on energy reduction, counterbalance rising power needs.
c. Studies overestimated datacentre power consumption; actual use is significantly lower.
d. Hyper-scalers like Amazon and Google can increase AI capacity massively without additional energy.
e. Datacentres not investing in efficiency are set to lose AI workloads to more efficient hyper-scalers.
4. Opportunities in edge AI (deploying AI at physical endpoints) over datacentre (cloud) AI:
a. Despite datacentre AI’s growth, edge AI presents better disruption opportunities.
b. NVIDIA dominates AI infrastructure, but edge AI’s unique challenges provide openings for new players.
c. Edge AI’s potential lies in diverse applications in health, automotive, and industrial sectors.
d. Edge AI’s focus on cost, energy-use, and privacy offers a vast opportunity, especially with low-power SoCs (systems on chips) and microcontrollers (MCUs).
e. Market disruptors in edge AI include Raspberry Pi (silicon), and Cambridge Mechatronics (interfaces).
5. Challenges in utilizing enterprise data for AI:
a. Despite advances, many enterprises still face obstacles in leveraging internal data for AI.
b. Enterprises transitioning from traditional to modern data stacks find the ecosystem difficult to navigate.
c. Therefore, most organizations are still at the ‘data reporting’ stage, aiming to become ‘data-driven’ with predictive analytics.
d. The ‘modern data stack’ (MDS) ecosystem complexity may lead to market consolidation and pushback from hyper-scalers.
e. GenAI introduces new MDS requirements (eg prompt engineering workflows), leading to opportunities for emerging startups and rising stars like UK’s Matillion.
