Morgan Stanley IM: Walled Gardens - The Slow Burn AI Winners

These “slow burners” will not only generate value from AI, but also have the pricing power to hang on to the resulting benefits. This challenge is easier in relatively closed systems, or “Walled Gardens,” that include proprietary data.

2023 has been an artificial intelligence (AI)-driven market, with the "Magnificent Seven"¹ dominating in the U.S., delivering over 85% of the S&P 500 Index’s returns this year. The "Seven" combined returned 43% in the first nine months of the year, as against 3% for the other 493.² Their success is arguably not surprising.

As we set out in our June 2023 Global Equity Observer “Compounding Through the Hype”, the early winners of the AI "gold rush" have been the shovel sellers, the semiconductor providers and the cloud "hyperscalers" who are responsible for the infrastructure necessary for generative AI deployment – specifically vast amounts of storage capacity and processing power. It is these companies that are already seeing the benefits of the new wave. The most extreme case is a Santa Clara-based American multinational technology company, whose forward earnings estimates have tripled this year, while on a less spectacular level the software and cloud computing provider we own has already claimed a 2% growth boost for Azure in the latest quarter. This early-stage revenue increase gives a line of sight to the likely significant increases in demand for the hyperscalers’ cloud services, though the revenue boost will be accompanied by significant increases in capex as they build the required capacity.

Alongside these clear beneficiaries today we believe there are others – the "slow burners" – for whom the benefits of generative AI (GenAI), and AI in general, will take longer to emerge, but could still be significant over time. These are more likely to be users of the models rather than the thought leaders, though they will of course be involved in creating use cases. These slow burners will need to be able to generate value for customers and/or reduce costs for themselves through GenAI, and most importantly have the pricing power to hang on to a decent chunk of the resulting benefits for their shareholders. It is leveraging the companies’ existing competitive advantages that brings the opportunities, using GenAI to add further value to their already excellent business models. By contrast, were GenAI to increase customer value or reduce producer cost in a commoditised industry it would be the customers that gained rather than the shareholders. Competitive pressures would force the companies to pass the fruits of the technology to customers, whether in unrewarded higher quality or lower prices.

Models and Walled Gardens

In May of this year an anonymously leaked memo, allegedly from a researcher at a global search engine company, made the claim that internally developed AI models have “no moat” when it comes to GenAI.³ This is because new open-source models, which are based on readily available application programming interfaces (APIs), are quicker, more adaptable, more private and, not to mention, free. Why would consumers pay for a restricted model when unrestricted alternatives compete on quality and price? A major player and multinational technology conglomerate has now made the code for its conversational AI freely available on the internet after it was leaked. The fact that the non-open-source players (and by implication their respective investors) are now arguing for government licensing of cutting-edge models suggests that they are feeling the competitive heat, given the U.S. tech sector’s usual antipathy to any form of government intervention.

Even if GenAI models do become relatively commoditised, the challenges in deploying them at a massive scale are far from trivial. Efficiency is key given the significant compute and memory costs in large language models, aside from the challenges around hallucinations, where the models simply make stuff up! Incorporating GenAI in search is about far more than the model. The challenge is far easier in relatively closed systems, or “Walled Gardens”, which is where most of the companies we are discussing come in. There are significant opportunities where companies have proprietary data within their garden, either pure or blended with more public data. Many have been using traditional, or predictive, AI on their data for years to generate insights or automation, and are now adding in the GenAI element.

GenAI and the Portfolio

One relatively early mover is an American credit rating agency we hold in our Global portfolios. Its analytics business has long used Predictive AI, for instance within its KYC (Know Your Customer) business to screen for red flags and to look for potential fraud. Having partnered with an American multinational software and cloud computing company we also own, it is bringing a ChatGPT-powered “research assistant” into the business to help customers navigate the system. The expectation is that this will bring significant efficiency gains for clients, as they can do their analyses far faster, with large elements of their investment reviews being written after a few prompts. It only uses the data within its own databases, and all statements will have sources attached which should limit the threat from hallucinations (and provide an extra revenue opportunity where clients don’t have entitlements to those sources). The charging model is still under review, but the company will look to “price behind value” – something it is pretty good at, as it already gets an extra 7% per year of revenue on average from existing clients on the back of cross-sells, upgrades and pricing.⁴ The company has also given its 14,000 staff the CoPilot app in a bid to generate ideas to improve the business. The current priority is the revenue opportunities, but there are significant expense gains to be reaped later.

There is a similar story for a U.S. financial data and software company we own. It too has introduced a GenAI interface to help customers interrogate its system or initiate tasks, and even help with Python programming, opening up its datasets to greater usage and creating customer value, helping retention and pricing. In the medium term there are also likely to be significant efficiencies from both client service – as GenAI helps handle client queries and content collection, the area where around 50% of its employees work – and as GenAI accelerates the acquisition and cleaning of the crucial data.

The opportunity does not just come where the company owns proprietary data. In the case of a European software company that is a leader in enterprise resource planning (ERP), the company effectively owns the building where its clients’ data is housed and analysed. As elsewhere, AI has been a key driver of analytics and automation for some time, already used by 26,000 business customers; for instance, intelligent collections in finance, cutting the time between invoice and payment, or predictive replenishment, automating reordering of materials.⁵ A GenAI CoPilot is now being introduced to allow its systems to be interrogated in natural language, driving analysis in finance, human resources (HR) or on the supply chain. GenAI can also be used to compose job descriptions and interview questions within the HR function or generate process models and documentation in the process transformation area. Alongside charging for point use cases, the company is planning to offer a GenAI version of its core public cloud product later this year at a 30% premium. Massive switching costs make it tough for clients to leave the company’s ecosystem. Strong use of AI will make staying more palatable and offer extra monetisation opportunities, before even considering the gains that GenAI may offer in more efficient coding, potentially accelerating the incoming margin gains from the ongoing cloud transition.

In all three cases the GenAI opportunity is evolutionary not revolutionary, offering an extra boost to the companies’ top-line growth and margin improvements on what are already successful, profitable growing businesses. There are many other examples of companies in the portfolio with AI-friendly franchises built around valuable data, for instance credit bureaus, professional publishers, insurance brokers or a health care data and clinical services provider. On top of this there are players in sectors such as consumer staples that have a significant edge on competitors as they have been investing significantly in their data, enabling an analytic edge. Accenture claims that only 10% of its clients are “data-mature” and able to fully exploit AI opportunities, meaning that those who have made the journey have a significant advantage.

Uncertain Times

The market has gone into reverse, dropping 7% over the last two months after the powerful recovery since September 2022.⁶ As on the way up, the decline has been driven by rating rather than earnings, which have remained roughly flat. The fall has come despite improving macro forecasts in the U.S., if not in Europe and China, though bears point to early warning signs, such as U.S. trucking employment and pending house sales. The catalyst for the market’s fall seems to have been the relentless rise in yields, with the U.S. 10-year approaching 4.6% at the end of September, up 73 basis points (bps) in the quarter and 46 bps in a month,⁵ even as the U.S. Federal Reserve (Fed) raising cycle comes to an end. There is plenty of speculation about the reason for this sharp rise in yields, be it higher rates for longer or better growth prospects, but it seems probable that it is simply about supply and demand. There is no shortage of supply of Treasuries, with the U.S. running a deficit of near 8% of gross domestic product despite sub-4% unemployment and $7.6 trillion of the existing stock due to mature in the next year,⁷ while there are real question marks about the demand appetite of two of the main historical buyers, China and the Fed.

These higher yields put two question marks over the equity market. The first is whether the massively indebted system can deal with far more expensive credit without something breaking, as was the case with U.K. liability-driven investment (LDI) strategies a year ago and Silicon Valley Bank in the spring. The second is the impact that the yields have on the relative valuation and attractiveness of the equity market versus bonds, as the gap between the market’s earnings yield and the “risk-free” rate has fallen to the lowest level in 20 years, a gap even lower than it was two months ago despite the fall in the equity market. Even ignoring the fixed income alternative, the MSCI World Index’s current 16.0x forward multiple does not look cheap, particularly as it is based on an arguably optimistic 10% earnings growth assumption for 2024 in what is likely to be a slowing economy, even if the authorities do manage to pull off a soft landing.⁵ It is difficult to argue that the market is embedding any significant chance of a downturn in either its multiples or earnings. Our thesis, as ever, is that pricing power and recurring revenue, two of the key criteria for inclusion in our portfolios, will once again show their worth if there is indeed a downturn, and the market would once again come to favour companies which have resilient earnings in a tough economy, making quality a relatively safe haven in these uncertain times.

¹ Meta, Apple, Nvidia, Amazon, Microsoft, Alphabet and Tesla
² Source: FactSet
³ Source: https://www.semianalysis.com/p/google-we-have-no-moat-and-neither
⁴ Source: Company financial reports
⁵ Source: Company financial reports
⁶ Source: FactSet
⁷ Source: Apollo Asset Management