Archived article: Please note that tax, investment, pension and ISA rules can change and the information and any views contained in this article may now be inaccurate.
Related
news
12 November 2025
Welcome to the new-look Shares magazine – why you should read it and what to expect1 min readAuthor
Shares Magazine
The world changed forever in November 2022 when private company OpenAI released ChatGPT, a generative artificial intelligence chatbot. This ‘large language model’ is a type of machine learning which can understand and generate natural language and images.
This means LLMs can generate human-like text, write code, create content and solve puzzles, while engaging in nuanced conversations, all through an accessible interface.
One of the unique features of LLMs is the vast computing power required to train the models on large datasets. This energy consumption aspect makes the technology very different from previous advancements.
The technology demands for AI training have been more than doubling every year according to analysts, and by 2030 an individual AI training could need four gigawatts of compute power.
Compute power refers to the hardware, processors, memory, storage and energy required to operate data centres. Research by business consultancy McKinsey shows that by 2030, data centres are projected to require $6.7 trillion in capital expenditures worldwide to keep pace with demand for compute power.
For context, a gigawatt is one billion watts and equivalent to powering hundreds of thousands of homes, or over 100 million modern LED (light emitting diode) bulbs.
The cost of building one gigawatt of datacentre capacity costs between $50 billion and $60 billion. Around $35 billion of that cost is needed fit out a data centre with Nvidia’s latest chips and systems.
Nvidia makes the fastest and most advanced AI GPUs (graphics processing units) which means the company supplies most of the computer chips which are powering the AI revolution.
Analysts at Mizuho Securities estimate Nvidia has a market share of between 70% and 95% for advanced AI chips. These attractive investment characteristics and the AI demand tailwind have helped to catapult Nvidia to a market capitalisation of nearly $5 trillion.
How fast is AI growing and how big could it be?
To give a sense of the growth momentum in ChatGPT, OpenAI’s revenue is forecast to have grown eight-fold from 2023 to around $13 billion at the end of 2025, making it one the fastest growing companies on record.
That sounds impressive, but when you consider OpenAI has converted less than 5% of its roughly 800 million weekly users into paying subscribers, the ongoing potential could be considerable.
The underlying generative AI market has seen annual revenues grow from $191 million in 2022 to $25.6 billion at the end of 2024, driven by customer demand and the entrance of competitor LLMs like Alphabet’s Gemini, Anthropic’s Claude, Elon Musk-owned xAI’s Grok and Meta Platforms’ Llama.
How big could the AI market become? Analysis by McKinsey suggests the total potential economic value from AI could reach $15.7 trillion by 2023, with generative AI contributing up to $4.4 trillion of that value.
Another perspective comes from Bain & Company, which estimates the total addressable market for AI hardware and software could reach between $780 billion and $990 billion by 2027.
What role is Microsoft playing?
Microsoft has invested more than $13 billion in OpenAI since 2019 and provides its Azure cloud services to the company.
OpenAI started life as a not-for-profit making company but recently announced plans to split itself into two pieces – a for-profit business to be called OpenAI Group PBC (public benefit company) and a not-for-profit company called OpenAI Foundation.
On 28 October the company announced Microsoft held a 27% stake in OpenAI Group PBC, valued at approximately $135 billion with the OpenAI Foundation holding a 26% stake. Meanwhile current and former employees held 47%. Prior to the recapitalisation and funding rounds Microsoft held a 32.5% stake in the for-profit business.
OpenAI raised $40 billion in March 2025, in the largest private funding round in history. The latest deal with Microsoft implies a current valuation for OpenAI of $500 billion.
As part of the agreement Microsoft said OpenAI had agreed to purchase a further $250 billion of services from Azure.
Microsoft’s exclusive intellectual property rights to OpenAI’s models have been extended to the sooner of 2032 or once AGI (artificial general intelligence) is announced and confirmed by an independent expert panel.
AGI is an AI system which rivals or exceeds human intelligence. Key evaluations include creativity, ethical judgement and abstract reasoning.
The agreement also allows OpenAI to develop some products with third parties, but these will be exclusive to Azure.
In an echo of the dot-com boom of the late 1990s, OpenAI is forecast to ‘burn’ though $8 billion of cash in 2025 as the company looks to secure chip capacity and develop further revenue sources.
The company is projected to lose $14 billion in 2026 with cumulative losses since 2023 expected to reach $44 billion by 2028. OpenAI has projected annual sales of up to $100 billion by 2029, driven by increasing AI adoption.
The Financial Times recently calculated that OpenAI has committed to take more than 26 gigawatts of computing power from Oracle, Nvidia, Advanced Micro Devices and Broadcom at a total rough cost of $1 trillion over the next decade.
A strategic deal with Nvidia centres on deploying huge amounts of Nvidia’s AI chips to run OpenAI’s next generation LLM models. To support this buildout, the AI chip maker intends to buy up to $100 billion of OpenAI’s non-voting shares, the proceeds of which will be used to purchase the chips.
Some analysts have questioned the circular nature of the deal which can be viewed as vendor financing. Such models were a feature of the dotcom boom in the late 1990s, but analysts point out a key difference is that today’s leading tech companies throw off lots of free cash flow.
While that may be true, the rising cost of AI infrastructure could result in lower free cash flows than investors currently assume, which is why it is noteworthy that debt has been used in some recent deals.
In August, Meta secured $29 billion from a group of private credit investors including Apollo Global, KKR, Brookfield, Carlyle and Pimco for its US-based AI data centre programme.
The complex arrangement keeps debt off Meta’s books which means it can continue to fund its AI ambitions. A joint venture will build and own the four million square-foot data centre while Meta will occupy and use the facility under a 20-year lease.
A similarly creative investment vehicle was created by xAI, comprising $12.5 billion in debt and $7.5 billion in equity, to fund the purchase of Nvidia chips that will be leased to xAI. Nvidia will contribute up to $2 billion of the equity tranche in another example of potential vendor financing.
What about the energy demands?
The power-hungry AI revolution has rekindled interest in nuclear power, including new technologies such as SMRs (small modular reactors), which as the name suggests are smaller than traditional nuclear reactors.
Their modular nature means the components can be made in a factory and shipped to a site for assembly, making them faster and more cost-effective than large scale reactors.
In September 2025, Microsoft signed a 20-year deal to secure nuclear power from Constellation Energy which plans to reopen the Three Mile Island nuclear facility in Pennsylvania. The site is known for the worst reactor accident in US history after a partial meltdown in 1979.
Google announced a ‘world first’ after inking an agreement with Kairos Power to buy energy from the SMRs the company is developing, set to come online in 2030.
Not wanting to be left out, online marketplace and cloud services provider Amazon recently signed three deals with nuclear energy suppliers including one with Energy Northwest in Washington, to develop four SMRs.
CEO of Amazon Web Services Matt Garman said: ‘One of the fastest ways to address climate change is by transitioning our society to carbon-free energy sources, and nuclear energy is both carbon-free and able to scale – which is why it’s an important area of investment for Amazon.’
It is worth pointing out that these deals are for power projects yet to be built using technologies which are unproven commercially.
OpenAI CEO Sam Altman has backed nuclear start-up Oklo, whose shares have risen six-fold so far in 2025, giving the company a market capitalisation of $22 billion.
According to S&P Global Market Intelligence, Oklo is the largest US company that generated no revenue in 2025. Another zero-revenue firm is Fermi which investors valued at $19 billion when it came to the market in September.
The company, backed by former energy secretary Rick Perry plans to build out 11 gigawatts of power for data centres, which, for context is roughly equivalent to the capacity of New Mexico.
Is the AI market a bubble?
A recent Bank of America fund management survey showed a record share of global managers believe AI stocks are in a bubble while Bloomberg reported that the number of news articles mentioning ‘tech’ and ‘bubble’ has spiked in recent weeks.
There is an additional worry that the big increase in AI infrastructure spending could be distorting the economy.
For example, analysts estimate US GDP (gross domestic product) growth would be less than 1%, compared with the current 3% reading, if AI infrastructure spending was excluded from the numbers.
The worry is that because growth is heavily concentrated in one sector, it could expose vulnerabilities for the wider economy if the AI boom proved unsustainable.
On the other side of the argument, Ben Rogoff, manager of the Polar Capital Technology Trust, is firmly in the camp that believes the AI story is real and not a bubble.
‘We feel that the size of the prize reflects the probably unparalleled opportunity associated with AI,’ explained Rogoff to Dan Coatsworth, Head of Markets at AJ Bell.
However, Rogoff concedes that for the first time in the evolving AI story, the case for some of the leading players deploying AI is less favourable compared with the companies providing the infrastructure.
While there has been clear adoption of AI by consumers as seen by the 800 million weekly users of ChatGPT, evidence that corporations are using the technology has been slower to emerge.
Nevertheless, Rogoff believes it is just a matter of time before companies start to appreciate the significant productivity gains to be had by adopting AI. ‘In our view AI will become a widespread, must own, general purpose technology,’ said Rogoff.
Which ETFs and trackers provide exposure to AI?
Trying to pick winners and losers from the AI field is a challenging prospect even for professional investors like Ben Rogoff. Therefore, investors interested in the sector may find it more prudent to examine passive trackers of technology-focused indices.
The largest specialist tracker of AI is the Xtrackers AI & Big Data ETF GBP, which has more than £5 billion of assets and an annual charge of 0.35%.
The fund seeks to track the Nasdaq Global Artificial Intelligence and Big Data index by fully replicating the constituents of the index. Dividends in the ETF are accumulated and reinvested back into the fund.
Top holdings include Alphabet, Nvidia, Apple, Oracle and Amazon, with US stocks comprising 81% of the fund’s assets and the top 10 holdings making up 45% of the total fund.
Another relevant ETF is the iShares Automation & Robotics UCITS ETF which is on the AJ Bell Favourite funds list. This is a shortlist of funds curated and researched by AJ Bell’s in-house investment experts.
The £3 billion fund seeks to replicate the performance of the STOXX Global Automation & Robotics index which tracks a broad spectrum of companies developing technologies in automation and robotics.
The index operates on an adjusted equal weight methodology whereby a cut-off is applied should the weighting concentrate the ETF too heavily towards smaller companies.
This product is well diversified across 170 names and has an annual charge of 0.4%.
Top holdings include chip makers Nvidia, Advanced Micro Devices, Intel, Japanese chip company Advantest, and Swiss electrification and robotics company ABB.
What about actively managed funds?
Investors can access funds which invest in an actively managed portfolio of AI companies. Investors should expect to pay higher annual fees for actively managed funds.
There are a range of open ended and closed ended funds available and some fund managers offer both versions for the same strategy.
The main difference is that closed ended funds, also called investment trusts, can trade a premium or discount to the value of the assets in the underlying portfolio.
Investment trusts can also borrow money to invest which is not possible for open ended funds. Investment trusts trade like shares while open ended funds are usually priced daily.
The largest investment trust investing in the technology sector is FTSE 100 constituent Polar Capital Technology trust, which has a strong pedigree as a specialist asset manager.
The £5.2 billion trust and £5.6 billion fund are steered by seasoned managers Nick Evans and Ben Rogoff and supported by a team of portfolio managers and analysts.
The managers look to add value by identifying inflection points in next-generation technologies and fundamental analysis of companies.
The resulting portfolio is well diversified and holds between 60 and 85 names. The fund version of the trust is on the AJ Bell Favourite funds list.
The fund is relatively expensive with an annual charge of 1.1%, reflecting its specialist nature. The investment trust trades at an 11% discount to net asset value.
The largest actively managed fund investing in the technology sector and related AI companies is the Allianz Global Artificial Intelligence Fund, which has nearly £6 billion of assets under management.
The fund is diversified across 54 holdings with US stocks representing 85% of total assets. Top holdings include chip makers Nvidia and TSMC, Tesla, Broadcom and Microsoft.
Martin Gamble: Shares and Markets Writer
Martin Gamble is Shares and Markets writer at AJ Bell. He was previously the Education Editor of Shares Magazine. He has been with the business since 2019.
Martin graduated from the University of Kent in...
