In the first eight months of 2024, Microsoft (MSFT), Meta (META), Google (GOOGL) and Amazon (AMZN) collectively recorded $125 billion in AI-related expenses (CapEx) and operating expenses, according to September. JPMorgan Report. The combined CapEx of these four technology giants alone is expected to rise past the 200 billion mark for the entire year of 2024.
AI startups, on the other hand, have received an unprecedented amount of money from investors eager to capitalize on the technology's profitable capabilities. OpenAI is expected to finish 2024 as the most funded AI company, recently valued at $157 billion. Rival Anthropic is preparing to raise $40 billion in new funding.
Flush with cash, leading AI companies are now tasked with proving to investors—and the public—that their pricey bets on the new technology will pay off. From the ongoing pivot to “agent AI” to new scaling rules and extensive testing of thousands of AI capabilities, here's a look at what 2025 will bring to the world of AI:
Agent AI will be the “next big breakthrough”
The buzzword refers to autonomous AI assistants that can complete tasks without human supervision. The potential of AI agents to improve workplaces and everyday life is quickly gaining traction in Silicon Valley, with companies like Salesforce touting agents as their next big product.
Microsoft, too, has introduced a number of AI agents in recent months. In November, it introduced several AI assistants designed for the Microsoft 365 suite, including an agent that can provide translation in nine different languages.
OpenAI is also on the “agent AI” train, and the upcoming model is expected to be able to perform tasks such as booking a trip and writing code. AI agents are “what's going to sound like the next big breakthrough,” said Sam Altman, CEO of OpenAI, during a recent AMA on Reddit.
The global market for AI agents is currently worth more than $5 billion, according to research firm MarketsandMarkets. By the end of the decade, this figure is expected to increase to $47 billion, driven in part by demand for agents among business customers.
A test-time computer could be the solution to the AI training data crisis
One of the key components of AI's success in recent years has been the large amount of data incorporated into AI models. But there is only a limited amount of text, images and videos on the Internet. To avoid a plateau in the development of the technology, AI companies are turning to other methods of training their models. One of the most promising solutions is to calculate test time, where AI models develop in consultation and take longer to think about possible answers before responding—a theory recently demonstrated by OpenAI's O1 model.
On an earnings call in November, Nvidia (NVDA) CEO Jensen Huang described the new OpenAI model as “one of the most exciting developments” in scaling and noted that “the longer it thinks, the better and higher-quality it produces.”
Huang is not alone in his hope. Microsoft CEO Satya Nadella also pointed to the calculation of test time as a new measurement rule in November, while OpenAI co-founder Ilya Sutskever earlier this month highlighted it as a continuation of the previous AI training period.
Artificial data is another promising solution
Another solution to AI's data problem is to replace traditional data with information generated by the technology itself. The synthetic data market is expected to increase to $2.1 billion by 2028, representing an increase of more than 450 percent from 2022, according to BCC Research.
Altman highlighted the power of synthetic data last year when he discussed the shrinking AI data supply, noting in an interview that “as long as you can get an event horizon for synthetic data, where the model is smart enough to make good synthetic data, I think it should be fine.” OpenAI, and competitors such as Anthropic, Meta, Microsoft and Google, have all reportedly started using artificial data in some way to train and tune models.
In October, AI startup Writer unveiled a new AI model trained entirely on AI-generated data. This method allowed the company to reduce the cost of building the model, which reached only $ 700,000 compared to the millions released by other companies. For example, OpenAI's GPT-4 model, costs more than $100 million to train.
“Big world models” will create 3D AI worlds
To date, most AI visuals remain two-dimensional, something technology pioneers are looking to change in the coming years. “Big world models” are an emerging form of AI that aims to create three-dimensional interactive scenes that enhance the worlds of movies, games and simulations.
One of the biggest players in this space is World Labs, a startup founded by Stanford AI pioneer Fei-Fei Li that raised $230 million earlier this year. The company is looking to build large-scale models of the world with “local intelligence,” the kind of intelligence that understands and interacts with the real world. To illustrate this concept, Li previously used the example of a cat trying to roll over a glass of milk and people's ability to predict the outcome of this event and therefore take action to prevent the glass from falling.
In early December, Google DeepMind presented its large-scale world model in the form of Genie 2, which simulates a virtual environment that will be used to train and test AI agents. The area will likely focus more on the lab going forward, as evidenced by its recent hiring of Tim Brooks, a former OpenAI researcher who oversaw its Sora video generator. In X's letter welcoming Brooks to his team, Google DeepMind CEO Demis Hassabis noted his excitement at “working together to make the longtime dream of a world simulator a reality.”
AI search engines will reshape internet search
Google has long been a seemingly untouchable force in the search market. But with the advent of AI, the rise of AI-powered search engines is looking to shake up the tech giant.
Not that Google hasn't embraced the technology itself. In 2024, it introduced AI Overview, a feature that provides users with AI-generated summaries rather than links. CEO Sundar Photosi predicts the feature will attract more than 1 billion users a month and is already “expanding search usage and user satisfaction,” he told Wall Street analysts in October.
But Google will have to contend with a crowded industry for search tools, as companies like OpenAI and Microsoft expand into the arena with AI help.
Meta is reportedly preparing to launch its AI-powered search engine, and startup Perplexity AI has emerged as a formidable player. Recently valued at $9 billion, its AI-powered search tools already process about 20 million queries every day, up from 2.5 million by early 2024.
