The Top 16 Artificial Intelligence Trends of 2025

AI models get smarter by being given more time (and tokens) to think

As developers allocate more computational resources — namely time and compute — AI models like GPT-4 and its successors have become significantly more accurate and contextually aware.

Edge AI: deploying AI on local devices for faster, cheaper processing and privacy

Edge AI, a combination of artificial intelligence and edge computing, enables AI processing directly on local devices, like smartphones and IoT sensors, bypassing the need for constant cloud connectivity. This shift to on-device processing reduces latency, lowers data transmission costs, and allows for more privacy by keeping sensitive data on the device.

The rapid growth in Edge AI is fueled in part by improvements in chip technology and increased concerns about data sovereignty and privacy regulations globally.

Generative AI explodes in popularity following ChatGPT’s launch

Since ChatGPT’s launch in 2023, millions of businesses have integrated generative AI into their customer-facing products and internal processes. Consumers, too, are using AI for everything from therapy to medical advice to help with schoolwork and everything in between.

The rise of AI companions surges alongside the rise of the loneliness epidemic and an aging population

Replika.ai, a site where people can befriend an AI chatbot, is often used for romance, but there’s a growing opportunity to build a massive business with a fascinating monetization loophole: AI companions for the elderly.

Recently, a number of companies have experimented with giving robotic pets to nursing home residents. Remarkably, one business called Paro got the FDA to classify their companion robots as a medical device.

This significant milestone allows the product to be eligible for insurance coverage and shifting the spend from the consumer to the insurance company means a whole lot more revenue potential and less price sensitivity.

Caretaking robots and AI companions have several significant advantages over humans. Unlike their human counterparts, they’re unlikely to ever become frustrated or impatient or forget doctors' appointments or reminders to take medication. And they won’t defraud or abuse their patients, a surprisingly common problem in the industry.

As AI becomes more capable, we can expect to see new players enter this untapped market of helping elders cope with the loneliness epidemic, one of many mental health trends.

Data labeling platforms on the rise as demand for high-quality, large-scale datasets rises

One of the challenges in AI is that data-labeling is a low-wage, labor-intensive job, while designing chips and algorithms is a very high-skill, high-wage one. Economies that can support one of these job categories often can’t support the other. China is an exception; one of the reasons their AI industry has done so well is that there’s a thriving tech sector in coastal cities, while provinces in the interior are often low-wage enough that data-labeling jobs make sense. Companies like DataAnnotation are taking this China-specific advantage, and spreading it around the world.

Amazon’s Mechanical Turk, originally built in-house by Amazon to help them annotate data, was one of the original sites to offer bite-sized gig work for people around the world. This work often involves training AI, by completing tasks like labeling images for video and Lidar

AI increasingly being used by consumers to design their tattoos

Shopify strategically built a business name idea generator website. It ranks #1 in search results for the popular query "business name generator" and helps them funnel users into their core offering.

Like entrepreneurship, many consumers say idea generation is the biggest hurdle to getting a tattoo; they can’t come up with or settle on a design. A company that lowers this barrier can not only funnel in more leads, but can also drastically expand the market.

With tattoos, many consumers today look to Upwork and Fiverr to get designs made from freelance artists. Now, some parlors are experimenting with using AI tattoo generators for user acquisition and to grow their market by lowering the barrier to entry, just as Shopify did with their business name generator.

And the market is already big: Nearly half of American adults have at least one tattoo and the percentage is growing.

Many consumers get tattoos that can be covered with clothing so the size of the market is actually larger than it appears. But even visible tattoos are rising in popularity, which is a self-fulfilling cycle: a visible tattoo is an advertisement that it's socially acceptable to have tattoos.

AI-generated, human-looking models seeing rising demand as way to better localize content and also to reduce legal liability

While it appears that the world’s most popular YouTube creator, Mr. Beast, has 131M followers, he actually has 23% more than that as he dubs all his videos in Spanish, French, and Portuguese under 6 separate YouTube accounts.

Localization – translating and optimizing for local markets – isn’t a new thing. Apple, for example, employs thousands of people focused solely on making sure their apps are properly translated and the user interface still works with the translations; a button’s size or layout might need to be changed if the translation is too long, for example.

Synthesia – a product that lets users create videos of talking human-looking models just from plain text – is increasingly helping brands localize their marketing materials with up to 120 languages. Imagine making a commercial that could instantly scale to hundreds of regions without the overhead of localizing each one individually.

This lets brands customize videos for any conceivable need or demographic. An advertiser, for example, can generate stock photos that mimic the appearance of their target demographic. In some markets, there isn’t enough diversity of local human models to train the AI to generate artificial ones.

Generated Photos, a similar company that creates images, not videos, also focuses on video game creation. It’s a big unlock to be able to add realistic diversity to background characters. Rather than painstakingly design dozens or hundreds of distinct characters, or limit the scope of the game, game designers can use automatically generated images to populate their worlds with a diverse cast of characters.

There’s also an interesting legal benefit of using AI models: Using photos of models, in particular, can lead to surprisingly complicated legal situations since every state has different laws around model release forms so AI-generated models help brands reduce liability concerns.

Perplexity continues to surge in popularity

In 2016, the number of zero-click Google searches – searches where the user got their answer from the top of the results page without needing to click any of the results – surpassed the number of normal searches.

This is a big deal and is a consequence of several factors. First, Google has gotten better at answering simple questions directly at the top of their results page. People are also searching more on mobile so they’re more inclined to be satisfied with the answer given there than doing additional research. 63% of Google search sessions are in fact on mobile as of 2022.

Most importantly, the nature of web search is that most searches are informational, as opposed to transactional or the starting point of a longer research journey; and this portion is growing: zero-click searches climbed to a staggering 65% in 2022. Google now sometimes even answers queries within autocomplete so users don’t even need to hit enter on queries like “how old is Obama”.

While Google is arguably a long way off from being outright replaced, its biggest vulnerability is here – with informational searches – and as informational searches grow, there’s even more surface area for new competitors to gain market share.

Perplexity.ai, one of the most popular AI search engines, was one of the first large-scale consumer web apps built atop GPT-3 to break through the noise and show strong signs of long-term usage.

And while many businesses built atop GPT don’t have as much inherent defensibility as their predecessors did, search is a space with uniquely strong economies of scale: the more searches, the easier it is to improve the quality of results.