Computing’s top tech predictions – 2025 edition

How will the tech sector change in the next 12 months?

New technology and a changing political landscape cloud the future, but Computing is here to sweep the uncertainty away.

Here are the Computing editorial team’s predictions for what the next 12 months will bring to the tech sector.

AI will shrink

Concerns have grown this year about the massive amount of training data AI language models need to perform their work. To plausibly answer questions ranging from what an election candidate stands for to how to make tomato soup, large language models (LLMs) need petabytes of information as a base.

But what about narrower use cases? What if you only care about using AI to spot cracks in concrete, or telling you about the latest moves in the IT industry? And what if your data is confidential?

Enter small language models (SLMs), which are trained on a much smaller (the clue is in the name) subset of data – say, images of cracks in concrete. These models bring AI to the edge of the network to empower the IoT, and can be tailored to a specific industry to meet niche requirements and recognise jargon.

Because most SLMs are built privately, for a specific company, the training data is kept in-house – so there are also fewer worries about confidentiality.

SLMs have started to take off this year, but in 2025 that growth will explode – especially as multiple SLMs can be used together across a tool chain.

IT leaders have also told us this year that they’re not seeing the return on investment they expect from AI, but vendors continue to plough money into the technology, so we predict SLMs will evolve quickly. On top of that, AI trials will move from experimentation to monetisation.

Digital carbon footprint will become easier to measure

The demands of genAI, and datacentre developments slated for the UK, are likely to drag datacentre energy consumption further into the public consciousness in 2025.

This is a good thing, because so far the only people that really concern themselves with this sort of thing tend to be tech or sustainability reporters and researchers (and possibly those opposed to having datacentres built in their neighbourhoods).

However, the Directive on Corporate Sustainability Reporting (CSRD) and updated Energy Efficiency Directive in the EU is likely to drive greater understanding among businesses of their digital carbon footprints. It’s not the same as reducing it, but it’s a start.

In the US there are changes due in climate disclosure reporting in California (where most tech giants are headquartered) and lots of initiatives from AWS and Microsoft to reduce embodied datacentre emissions. Examples include a timber datacentre pilot from Microsoft and a bigger focus on reuse and refurbishment from AWS.

Smaller datacentre industry players are also focusing on sustainability, and there are specialised technologists like Deep Green and atNorth, who are innovating to reuse the heat generated by datacentres and channel it where it is needed.

These initiatives are all happening because businesses and financiers are showing more interest in the environmental impact of digitisation. Whilst datacentre emissions are not going to fall in the short or even medium term, they will probably grow more slowly than they would have done had businesses not been encouraged by governments and regulators to try and measure the impact of their digitisation.

If this continues to shape the strategy of data intensive businesses and the cloud/datacentre industry in 2025, so much the better.

Arrival of a long-promised Bill

This is one that doesn’t require a crystal ball, since it’s been slated for early 2025: the long-awaited UK AI Bill will be introduced.

Tea leaves are also not required to say that the UK will seek to position itself in its traditional comfort zone between the prescriptive, risk-based EU approach and...whatever comes out of the US.

That last point has me (and perhaps Whitehall officials) reaching for the magic runes, because who knows what that will be? The instinct of the Republicans will be to deregulate, but business is looking for certainty around AI, and laws provide that. Trump may decide to play favoured companies and countries off against less favoured ones with favourites changing on a whim depending on their perceived fealty.

The runes tell me the UK will cleave closer to the larger and more predicable EU market, while leaving as many options open as possible – but I got them cheap from Runes4U and can’t vouch for their reliability.

Big Tech breakups take a backseat

Sticking in the US, the new administration will likely roll back moves to break up tech giants like Apple and Google under antitrust law, if for no other reason than the measures were advanced under the Biden administration. Trump has already said China is afraid of Google, which would be another reason.

Already, moves have been made against attempts to mandate net neutrality among the carriers, so that would be more of the same. Big Tech will become even bigger.

AI will pop up in PC refresh cycles

With a standard laptop/desktop refresh cycle averaging 3-5 years, and most businesses having taken the opportunity to perform their last mass refresh in the pre-Windows 11 pandemic years, 2025 heralds the start of the next big cycle. And this time, it’s going to be all about the AI PC.

AI PCs are those that can run AI processes natively in the background, without impacting the normal run rate. While Microsoft would have you thinking these devices need a Copilot key, really the only requirement is the inclusion of a neural processing unit (NPU).

IT leaders’ opinions on AI’s usability are still divided, and we don’t expect AI PCs to drive a refresh by themselves – meaning, most users won’t replace a perfectly good PC just to get AI functionality. But, many IT leaders are likely to future-proof their estate and choose a PC with AI functionality when the inevitable refresh cycle starts, especially as operating systems and apps start to leverage the NPU.

For now, AI PCs remain premium devices, so don’t expect enterprise-wide adoption anywhere except a few niche cases; it’s more likely we’ll see a few teams getting these newest PCs and then spreading to the rest of the company as prices come down.

DEI will evolve – quietly

2024 hasn’t been a good year for DEI. “Go woke, go broke” has been the motto of activists who have targeted companies like Bud Lite and Disney for their publicly progressive policies.

Right-wing critics have succeeded in misrepresenting DEI as being responsible for everything from a bridge collapsing in Baltimore to the Crowdstrike outage and the attempt on Donald Trump’s life in the summer. The latter won the US popular vote whilst framing Kamala Harris as a “diversity hire.”

Those wilfully distorting and misrepresenting DEI have been ably assisted by the ranks of the noisiest enthusiasts for identity politics and lazy corporate DEI woke washing.

So, will 2025 be the year that DEI does the decent thing and just dies?

I’m not so sure. If the current backlash against DEI shames some of the most egregious corporate examples of woke washing, that isn’t a bad thing. DEI as marketing doesn’t make organisations more diverse, equitable or inclusive. Neither does forcing everyone to do unconscious bias training and declare their pronouns on their email signatures.

More sensible tech organisations will continue to quietly “do the work” on DEI. Many work hard to find talent from less privileged socio-economic backgrounds, and are working to implement the kind of family friendly policies that enhance employee retention.

Bjg Tech might be bellowing RTO mandates, but most tech employers will continue to quietly run a hybrid model because it suits them and their employees,

The DEI industry has colluded in its own failure by overreaching and losing sight of its original goals. There are lots of ways to build a diverse, equitable and inclusive workforce. It needs executive support, sponsorship, a moderate tone, time and an abundance of good faith. More light, less heat.

DEI will not die in 2025. It will quietly evolve and is likely to be more successful as a result.