investment insights

    AI’s infrastructure points to a first phase of monetising

    AI’s infrastructure points to a first phase of monetising
    Marco Barresi - Senior Equity Research Analyst, Tech sector

    Marco Barresi

    Senior Equity Research Analyst, Tech sector

    Key takeaways

    • Semiconductors have become strategically important and the industry is attracting government subsidies and tax credits globally
    • AI’s evolution depends on investments in semiconductor infrastructure. Adoption remains early stage, making it difficult to foresee which business models will succeed
    • AI will create a new generation of tech start-ups and applications. A second wave of monetisation will see AI packaged as a service to enhance existing software products
    • We like chipmakers exposed to AI developments through the cloud. As the economic cycle evolves, we continue to focus on quality technology companies.

    The semiconductor industry is already of strategic importance. The development of artificial intelligence (AI), which depends on increasingly sophisticated computer chips, is only intensifying the stakes. As AI creates new and essential tools, it will become a core component of a modern economy, company and careers. We look at the early steps underway to monetising generative AI.

    AI may have only surfaced as a public debate relatively suddenly, but the technologies behind it – the internet, smartphones and powerful computer chips - have taken decades to develop. AI’s first iterations have been mostly about replacing or accelerating tasks such as customer service chat bots, automatic translation, or improving search engines. But as generative AI becomes more capable, tools promise to augment human learning and productivity, applying massive data across myriad industries while lowering the threshold for creativity by putting new computer code in the hands of everyone. New consumer applications are likely in computer vision, natural language understanding, as well as tools to accelerate coding itself, and other applications that let cars or robots make automated decisions.

    While there is plenty of discussion around AI’s promises and threats, adoption among enterprises is still at an experimental stage. That leaves investors unclear which business models will succeed, and little visibility on how corporations can monetise the technology. This said, the pace of adoption can only accelerate as exceptionally high levels of interest translate into new solutions.

    Last month, Nvidia Corporation, a US-based computer chip maker, surprised markets with sales and earnings guidance into 2024 that was much higher than expected. As a result, its market capitalisation approached USD 1 trillion. Historically, Nvidia has focused on computer gaming, but the firm’s production of highly sophisticated chips that are central to building AI models in data centres has seen its market cap jump to just behind Amazon, and ahead of Meta (formerly Facebook).

    AI depends on training and inference... These computing-intensive tasks depend on GPUs

    Read also: Terminated or turbocharged? Analysing AI’s potential impact on jobs, industries and growth

    At first this looks at odds with the broad market for semiconductors. High inventories and slowing consumer demand for computers and mobile phones has weakened the market in 2023 with an oversupply of less sophisticated computer memory chips. However, the story for the GPUs or graphics processing units that Nvidia specialises in – with around four-fifths of the market - is very different. Specialist GPU ‘cores’ manage many memory-intensive computer tasks in parallel and were, until recently, mostly applied to gaming or video editing. The development of generative AI depends on training and inference for its machine-learning algorithms. These computing-intensive tasks depend on GPUs, and development should continue to drive semiconductor advances for calculations, storage and data transmission.


    Energy and second-stage services

    Once machine learning models have been ‘trained,’ the bulk of the data processing workload carried out by AI applications can depend on less-costly, and more power-efficient microprocessors. Data centres are hugely energy intensive, consuming an estimated 1% of global power. Massive energy demands means this lengthy AI training process is also driving efforts among cloud providers to improve the hardware running data centres.

    Nvidia is an illustration of the prerequisite infrastructure that makes AI applications feasible, and the first wave of value creation in the technology. We believe that a second wave will see generative AI sold as a service, packaged with software products and in enhanced versions of existing applications running on a personal computer and smartphones, for which the provider will charge a premium.

    A second wave will see generative AI sold as a service

    As that happens, we will see more regulation. There are a multitude of concerns, from job substitution, or ‘creative destruction,’ that are beyond the scope of this article. Nevertheless, AI faces headwinds in the form of trust and accuracy, problems around intellectual property protection, and data bias. New legislation will have to manage one of AI’s inherent weakness, which is that it lacks an audit trail to demonstrate the factors that led to its output. That may not have life-or-death implications for creative work such as film scripts, but has liability considerations for self-driving vehicles or medical applications.

    Read also: Taking stock of China’s reopening, and prospects

    Regulation will also have to tackle what is known as ‘AI alignment,’ or the responsibility of ensuring that generative AI matches human values to, for example, prevent models from answering potentially harmful questions. It is impossible to avoid the discussions right now around the potential that these tools represent to be weaponised, or turned to cyber-attacks and dis-information campaigns. A first step was taken at the Group of Seven meeting in May when leaders agreed to regular working group through what they labelled the ‘Hiroshima AI Process,’ referring to the G7 summit’s venue.

    Read also: IT’ll be back

    Investment and strategic concerns

    The G7’s project is a reminder that semiconductors are not only a global industry but have become a strategic concern to governments. Nowhere is that clearer than the subsidies legislated over the last two years to build capacity. These semiconductor ‘nationalisation’ programmes will drive growth, competition and innovation in the industry where a state-of-the art fabrication plant takes some two years and costs as much as USD 12 billion to construct. The complexities don’t stop there. Producing semiconductor chips involve multiple steps and there is a limited pool of scientists able to work on, and develop, the technologies.

    The US’s Chips Act, passed in August 2022, will provide USD 280 billion in subsidies over the next decade, including some USD 39 billion designed to boost domestic semiconductor production as well as tax credits worth another USD 24 billion. In the European Union, spending approved in April 2023 provides incentives for public and private investments in manufacturing facilities for chipmakers and their suppliers worth an estimated USD 47 billion. Of this, USD 33 billion will be directed towards new plant construction. Japan, South Korea and Taiwan also all provide tax credits or subsidies.

    Generative AI will breed the next generation of start-ups…

    In the meantime, China is working on support that is worth as much as the US, EU and Japan combined; an estimated USD 143 billion of subsidies and credits to its semiconductor industry over five years. US efforts to prevent semiconductor manufacturers from selling their most sophisticated chips to China may force the Chinese industry to innovate, creating solutions that in the long run, may accelerate its productivity.

    Despite the ethical, geopolitical and technological challenges ahead, we believe that generative AI will breed the next generation of start-ups, much as the arrival of the iPhone built an entire industry around mobile applications, and the rise of cloud computing created a new sector of software companies.

    For investors, the global semiconductor industry’s revenues were worth USD 574 billion in 2022. Of this total, 31% were in the most sophisticated computing chips. In the short term, lower demand for less-sophisticated chips is only slowly reducing high inventories. We prefer semiconductor producers serving the cloud market, and so exposed to developments in AI, or electrification. This fits well with our general preference for quality technology companies, as the economic cycle evolves.

    Important information

    This is a marketing communication issued by Bank Lombard Odier & Co Ltd (hereinafter “Lombard Odier”).
    It is not intended for distribution, publication, or use in any jurisdiction where such distribution, publication, or use would be unlawful, nor is it aimed at any person or entity to whom it would be unlawful to address such a marketing communication.
    Read more.


    let's talk.