AI’s back – and it’s just getting started

Please note: This article is more technical in nature than our typical articles, and may require some background knowledge and experience in investing to understand the themes that we explore below.

All data referenced in this article is sourced from Bloomberg unless otherwise stated, and is accurate at the time of publishing.

The first half of 2025 has shown just how easily markets can be swept up in a single narrative. For most of 2024, artificial intelligence (AI) was front and centre – dominating the news, drawing in investor interest, and helping to fuel a surge in tech stocks.

But more recently, it’s taken a back seat, as tariffs and talk of trade wars have drowned out almost everything else. Even so, AI hasn’t disappeared. Innovation hasn’t paused. 

And now, with ‘peak’ trade noise hopefully behind us – and six months on from the arrival of DeepSeek R1, a cutting-edge language model developed in China – AI could be ready to step back into the spotlight.

The new AI frontier

AI models aren’t just improving; they’re now surpassing human performance in some seriously tough areas – from coding challenges to solving scientific problems once reserved for PhD-level experts¹.

At the same time, the focus is shifting. Instead of just focusing on building ever-bigger models², researchers are also starting to prioritise smaller, faster and more sustainable ones that still deliver top results. DeepSeek is at the heart of this shift. Its use of reinforcement learning³ to strengthen reasoning capabilities, plus a commitment to open-source development, has already helped to reshape the AI space.

A flaw in the system?

AI might be getting smarter – but it’s also getting stranger. As it grows more powerful, it’s starting to feel a little like something out of The Matrix – a system that sounds convincing, but one that sometimes spits out confident answers that simply aren’t true.

In the industry, it’s called a ‘hallucination’ – when AI models make things up. An OpenAI report found that nearly half of the queries to its latest models were affected⁴. Third-party testing of its o3 model even showed it sometimes invented steps it claimed to have taken to reach an answer.

The worrying part? No one really knows why it’s happening – or why it seems to be getting worse as models grow more advanced.

Still top of mind

Tariffs may have dominated first-quarter earnings calls – appearing in 43% of them – but AI wasn’t far behind, coming up in 39%⁵. It’s clear the topic still matters to businesses.

Mentions of ‘agentic AI’ – models that can use other software and act on their own – jumped 275% from the previous quarter, underlining rising interest in advanced applications that could drive productivity.

Despite some well-known challenges, AI is becoming embedded across industries and daily life. Meta, for example, reported a 30% rise in advertisers using its AI-powered creative tools in just the first three months of the year.

Meanwhile, paid subscriptions to large language models among US businesses have climbed significantly – supporting the idea that we’re only at the beginning of a broader wave of AI-driven productivity.

Surveys also show that usage is no longer experimental. For many businesses, AI is already part of normal day-to-day operations⁶.

Beyond Silicon Valley

The US still leads the pack with cutting-edge AI models, driven by a powerful ecosystem which includes Google, Microsoft, Meta, Amazon, IBM and OpenAI. Yet China is rapidly closing the gap. For instance, DeepSeek’s R1 seemingly matches the performance of US models but at a much lower cost, underscoring the swift pace of competition.

China is also now nearly on a par with the US in terms of key AI benchmarks and leads the world in AI research and patents. Plus, AI launches in the Middle East, Latin America and Southeast Asia mean this race is going global.

California-based Nvidia’s near-monopoly on AI hardware is also under pressure as it faces up to some serious challengers. Many customers are now building their own AI-specific chips⁷ or teaming up to develop software that can rival Nvidia’s dominance, opening the door for alternatives.

In April, Nvidia fired back with NVLink Fusion, a tech that boosts chip-to-chip communication and lets partners build their own custom AI setups. Meanwhile, AMD – another major US chipmaker – made waves with a $10 billion Saudi-backed push into AI. Whether that’s enough to loosen Nvidia’s grip is still unclear.

But the real fight is over power – and price. OpenAI has a massive first-mover advantage, but as ‘agentic’ AI grows and businesses look to scale efficiently, the potential for huge market growth is hard to ignore.

AI momentum builds

After a rocky start to the year, AI stocks have bounced back, but valuations are still lower than six months ago (despite the wider market recovery). A broad basket of AI-related companies⁸ currently trades at 24.7 times forward earnings – below their five-year average and close to the 10-year trend⁹.

Yet the real story lies in the spending. The big four cloud service providers – Amazon, Alphabet, Meta and Microsoft – are set to increase AI-related capex and R&D by 35% in 2025, to nearly $600 billion – more than half of all US equity capex growth.

The scale is staggering, and so is the demand. Microsoft processed five times more AI inputs this quarter than last year – a key AI workload measure – with usage surging in March alone. AI revenues are racing ahead, but can this pace be maintained?

But the big question now is: who stands to benefit most from AI’s next phase? Could the spotlight move beyond the tech giants as enterprise adoption spreads? Is the focus shifting from enablers to users? And which businesses could benefit next?

What next?

The hype may have cooled since ChatGPT-4’s debut, but the structural shift AI represents looks to be just getting started – even as worries over hallucinations and made-up facts persist. Global investment, accelerating usage and rapid innovation all suggest the market is still in its early stages – with momentum building and a theme that’s increasingly hard to ignore.

Related articles