5 AI Trends To Look Out For in 2026

Four years after ChatGPT's first release, it's clear that AI is impacting our daily lives and reshaping the global economy.

But what is happening right now, in 2026, and where is all this heading?

In this post, I'm going to look at some of the most important trends that are currently dominating the world of AI.

If you like this post, you might also be interested in our roundup of AI statistics.

Trend 1: Agentic AI nears takeoff

The shift to agentic AI can be understood as the shift from using AI to answer our questions to using AI to perform actions.

This "agentic leap" is a big deal because it means that AI tools can start to orchestrate complex workflows (with a human in the loop), which will massively expand the potential use cases for AI in business.

Imagine this scenario: a company is currently using a generative AI chatbot to draft replies to a customer support question about being charged twice for a product they bought online.

The shift to agentic AI in this scenario could involve the AI autonomously confirming the double charge in the company's billing system, issuing a refund through the company's payments API, and then drafting an explanation for the customer, which it will send once it has human confirmation.

As you can see, the potential usefulness to a business of the second scenario is vastly greater than the first, although that is assuming that it works well.

At Synthesia we offer two agentic AI tools. We have our interactive video agents, and we also have our agentic AI assistant, which helps you to create a video and can be seen in action in the video above.

Trend 2: A surge in AI-driven training and upskilling

It's common to see AI and job disruption and losses discussed together.

That's not really surprising when you have AI leaders like Dario Amodei (CEO of Anthropic) warning that AI will wipe out half of all entry-level white-collar jobs.

Is that just a doom-marketing strategy for his company? Maybe.

But some more sober analysis from the World Economic Forum (WEF) does agree that there will be job disruption.

The WEF expects that 39% of existing skill sets will become outdated between now and 2030 due to AI-driven change, and that 59% of the global workforce will need to be either upskilled or reskilled by 2030.

The report goes on to estimate that 11% of this workforce is deemed unlikely to receive this needed retraining, which implies that there are over 120 million workers at risk of redundancy.

That's the challenge facing businesses — getting their workforces ready for a very different AI-powered world. But it appears that many companies and governments are facing up to the challenge.

The WEF found that 77% of employers surveyed plan to upskill their workers in response to AI-driven change.

Here at Synthesia we are planning to work with businesses to meet this challenge with our upcoming conversational agent product that will help train and upskill employees.

We also recently signed an MoU with the UK Government’s Department for Science, Innovation and Technology (DSIT) to advance AI adoption and upskilling across the UK.

Trend 3: The continued infrastructure buildout

If the demand for AI is going to continue to grow massively, then that'll need a massive growth in compute to power it.

That's the logic behind a massive wave of AI-related infrastructure investment that Morgan Stanley estimates at nearly $3 trillion by 2028, with more than 80% of that spend still to come.

What does AI-related infrastructure actually mean? It includes:

• Power and energy investments: Energy is likely to be the biggest bottleneck in the future as our power generation and grids can't currently cope with the projected demand. Nuclear and gas are probably the most common sources of AI-driven planned new power generation capacity.
• Compute hardware: GPUs, TPUs, high-bandwidth memory, fabs, and advanced packaging.
• Data centers: This refers to the actual buildings themselves as well as the land.
• Networking and interconnect: All the high-speed fabric that we'll need to link up all the chips and data centers.

These investment flows are so massive that they are having an impact at the macroeconomic level, with huge supply chain bottlenecks at various stages of the buildout causing inflationary pressures. A recent example was Apple raising the prices of their MacBooks and iPads as memory costs skyrocketed due to AI-driven demand.

Then there's the question of who is financing all of this? Big tech companies are cutting back on stock buybacks so that they can invest more in AI infrastructure, and private credit markets are having a growing role in the financing of data centers.

And what if the projections are wrong? What if AI demand is being overestimated? What if it's being underestimated? If a new model drops that changes everything, or if agentic AI takes off, then all these demand forecasts might have to be changed with potentially huge ramifications for the economy and financial markets.

Trend 4: AI taking a bigger role in scientific discovery

AI is increasingly playing an important part in scientific discovery.

Research teams across a wide variety of disciplines are using AI to generate hypotheses, control experiments, and collaborate with colleagues.

A recent real-world example of this was OpenAI's breakthrough on the planar unit distance problem, which is a famous open question in mathematics that was first posed by Paul Erdős in 1946.

It was the first time that an AI model autonomously solved a prominent open problem in mathematics, so it represents a pretty incredible milestone.

What makes it even more remarkable is that the proof came from a general-purpose reasoning model rather than a model that is specifically designed to solve this type (or any type) of math problem.

It's pretty amazing to think about what other scientific problems AI might be able to solve in the near future as it gets better at performing long and complex chains of reasoning.

Trend 5: A shift to smaller and more specialized models

With concerns about the ROI of enterprise AI spend gaining some traction (for example, at Uber), it feels like the "bigger is better" era of large general models might be about to give way to smaller and more specialized models.

It's not just about cost, though — the other factors pushing the market in this direction include latency concerns as well as data-sovereignty needs, the latter of which is an increasingly hot topic in geopolitical circles.

The idea is that it might be better to use smaller multimodal reasoning models that you can easily tune for whatever specific domain you want, as these could in theory provide similar levels of performance and accuracy within that domain as the larger general models that cost billions of dollars to build, train, and run.

Without the huge costs of these larger models, which get passed through to the enterprise end users paying token-based pricing, the economics could allow for a much greater use of AI in a wider variety of applications and devices.