6 Investor Predictions For AI In 2025

Forbes

14-05-2025

Investment and analysis money cash profits. Successful investor or entrepreneur making investing ... More plans. Financial consulting, investment and savings. Vector illustration.
That are we likely to see as we continue to integrate the results of LLMs into the business world?
There are certain overarching trends and patterns that experts are pointing to in gaming out the next few years of the AI revolution. For example, there's the tremendous need for data (which we'll talk about a little later). There's the need for energy, which is prompting new interest in nuclear power, and other means of sourcing the juice needed for new data centers and systems.
Then there are other sorts of changes getting recognized as investors prepare to take on the rest of 2025, and build strategy for successive years.
In a recent presentation at Imagination in Action, Ulrike Hoffmann-Burchardi, the CIO for Global Equities at the UBS Chief Investment Office, pointed to some interesting predictions for the AI market.
The first of her predictions was that AI data center spending would hit $4 billion by 2027, and exceed spending on other general purpose data centers. She listed three reasons. One relates to Jevin's paradox, where innovations mean more of the new technology gets used. Another contemplates training and scaling laws in the industry. The third centers around inference.
'We're not just doing one-shot inferencing,' she said. 'We are now simulating different decision paths, and trying to find out what the best solution is going to be - that takes a lot more compute.'
Another of Hoffmann-Burchardi's predictions is that, in the race to feed AI engines, synthetic data will help. That's going to help address the problem of needing original data to give systems something to train on. It also alleviates some issues around intellectual property and data ownership that come with using human data for AI ends.
Hoffmann-Burchardi hinted at continuing gains for AI hardware, while cautioning that some forms of hardware have incredible staying power, like the GPU, that has remained a popular format for over a decade.
This corresponds to some of Hoffmann-Burchardi's writing in a report in February where she talked about hardware trends.
'The release of Groq 3 by xAI on 17 February marked another proof point of hardware scaling laws,' she wrote. 'Groq 3 was trained on 100,000 NVIDIA H100 GPUs and, based on xAI's claims, it surpasses leading models from OpenAI and DeepSeek in areas such as mathematics, science, and coding, while it already achieved top rankings on the HuggingFace chatbot leaderboard. The capex spent by xAI for the development of its models confirms our view that the AI race will keep pushing the frontier of hardware spending.'
In other analysis, Hoffmann-Burchardi suggested that AI margins will actually go up.
'The reason for that is that these companies are their own best customers,' Hoffmann-Burchardi said. 'They use AI in order to lower costs and increase efficiency.'
As ballast, she pointed to news from a recent earnings call where a big company saved $260 million, or 2.6% of net income, by using GenAI to upgrade 30,000 job applications.
Hoffmann-Burchardi also cited Google estimating that 25% of their codebase is now written by AI. That will increase, she suggested.
'You'll see more and more of that,' she said.
Going back to Hoffmann-Burchardi's February report on global equity strategy, we have the assertion that China will get a significant amount of market share for Internet stocks:
'China's internet companies—from Alibaba, Baidu, ByteDance, to Tencent—have shown an impressive cadence of algorithmic innovation in developing efficient large language models since 2023. These models can be used to enhance the user experience, improve ad targeting, and personalize content, potentially leading to higher revenue. Further automation through AI agents can drive operational costs lower, setting up for an inflection in operating margins.'
Responding to questions from the audience, Hoffmann-Burchardi talked about two options – decreasing cost by letting people off, and improving revenue. If companies can find ways to improve revenue, she suggested, they won't just use AI for productivity – they'll find completely new ways of doing business, and that may allow people to keep their jobs.
'GPUs are ideally suited for parallel processing, especially if the algorithm, the transformer algorithm, is starting to evolve. Yes, there will be some competition (but) we believe (that) five years from now, GPUs will still be the dominant AI chip.'
'We actually think any industry, any company can re-invent itself for AI. You need just two things. You need an AI-first mindset. You need some proprietary data, … any company will have a chance to become a leader in its space.'
'AI is reinventing the way business is done at a speed that is accelerating, and that may be even difficult for us to fully understand.'
Those are some of the investing predictions around the AI world this year. Look for more as we continue to demonstrate what's coming out of MIT conferences and events, and other sources like professional blogs by AI experts, as well as the biggest news stories of the year.