In the first eight months of 2024, tech giants Microsoft, Meta, Google, and Amazon collectively reported an astonishing $125 billion in expenditures related to artificial intelligence (A.I.), encompassing both capital investments and operational costs, as revealed in a September report by JPMorgan. This combined capital expenditure from these four companies is projected to exceed $200 billion by the end of the year.

In parallel, A.I. startups have been attracting record amounts of investment, with many eager to capitalize on the technology’s promising future. OpenAI is anticipated to conclude 2024 as the most financially supported A.I. firm, recently valued at $157 billion, while its competitor Anthropic is preparing for new funding that could elevate its valuation to $40 billion.

With significant financial backing, these leading A.I. firms now face the critical challenge of demonstrating to both investors and the public that their substantial investments in this emerging technology will yield returns. As the industry transitions toward “agentic A.I.” and explores new scaling methodologies, let’s delve into what 2025 may hold for the A.I. landscape:

Agentic A.I. is set to be “the next major breakthrough”

This term refers to autonomous A.I. assistants capable of performing tasks independently of human intervention. The potential of A.I. agents to transform workplaces and daily life has garnered significant attention in Silicon Valley, with companies such as Salesforce positioning agents as their next flagship offerings.

Microsoft has also introduced a range of A.I. agents recently. In November, they launched several A.I. assistants tailored for the Microsoft 365 suite, including one that can translate text in nine languages.

OpenAI is similarly entering the “agentic A.I.” space, with an upcoming model expected to handle tasks like travel bookings and code writing. Sam Altman, CEO of OpenAI, stated in a recent Reddit AMA that A.I. agents are poised to represent “the next significant breakthrough.”

The global market for A.I. agents is currently valued at over $5 billion, according to MarketsandMarkets. By the decade’s end, this figure could skyrocket to $47 billion, largely driven by demand from enterprise clients.

Test-time compute may address A.I.’s training data shortage

A major factor behind A.I.’s success in recent years has been the vast amounts of data fed into its models. However, the availability of text, images, and videos online is limited. To prevent stagnation in technological advancement, A.I. companies are seeking alternative training methods. One promising approach is test-time compute, where A.I. models enhance their performance by reasoning and taking additional time to consider potential responses—a concept recently showcased by OpenAI’s o1 model.

During a November earnings call, Nvidia CEO Jensen Huang referred to OpenAI’s new model as “one of the most thrilling advancements” in scaling, noting that “the longer it deliberates, the higher the quality of the response.”

Huang’s enthusiasm is echoed by Microsoft CEO Satya Nadella, who identified test-time compute as a new scaling methodology in November. OpenAI co-founder Ilya Sutskever also emphasized its importance as a shift from A.I.’s pre-training phase earlier this month.

Synthetic data presents another viable solution

Another strategy to tackle A.I.’s data dilemma involves substituting traditional datasets with information generated by A.I. itself. The synthetic data market is forecasted to soar to $2.1 billion by 2028, marking a more than 450% increase from 2022, as per BCC Research.

Altman hinted at synthetic data’s potential last year while addressing A.I.’s dwindling data resources, suggesting that “once you surpass the synthetic data event horizon, where the model can generate high-quality synthetic data, things should fall into place.” Major players like OpenAI, Anthropic, Meta, Microsoft, and Google are reportedly already integrating synthetic data into their model training processes.

In October, the A.I. startup Writer introduced a new model trained entirely using A.I.-generated data, which significantly reduced development costs to just $700,000, compared to the millions usually spent by other firms. In contrast, OpenAI’s GPT-4 model required over $100 million for its training.

“Large world models” will revolutionize 3D A.I. environments

So far, much of A.I.’s visual output has been two-dimensional, but innovators are aiming to change that in the coming years. “Large world models” are a new type of A.I. designed to create interactive three-dimensional environments, enhancing experiences in movies, gaming, and simulations.

A significant player in this domain is World Labs, a startup founded by Stanford A.I. expert Fei-Fei Li, which raised $230 million earlier this year to develop large world models with “spatial intelligence”—the ability to comprehend and interact with the physical world. Li has previously illustrated this concept by describing a cat knocking over a glass of milk, emphasizing how humans can foresee and act to prevent such outcomes.

At the beginning of December, Google DeepMind introduced its own large world model called Genie 2, designed to simulate virtual environments for training A.I. agents. This area is anticipated to be a primary focus for the lab moving forward, as evidenced by their recent recruitment of Tim Brooks, a former OpenAI researcher who oversaw its video generator, Sora. In an X post welcoming Brooks, Google DeepMind CEO Demis Hassabis expressed his enthusiasm for “collaborating to realize the long-held dream of a world simulator.”

A.I. search engines are set to transform how we search online

Google has long enjoyed a seemingly unassailable position in the search engine market. However, the rise of A.I. could herald a shift in this landscape, as a new wave of AI-powered search engines emerges, poised to challenge Google’s supremacy.

Google itself has embraced A.I. technologies, launching AI Overviews in 2024, which provide users with A.I.-generated summaries instead of standard search links. CEO Sundar Pichai is optimistic about this feature, predicting it could attract over 1 billion monthly users and enhance search engagement and satisfaction.

Yet, Google faces fierce competition from other search engine innovators like OpenAI and Microsoft, both of which are enhancing their offerings through A.I. Meta is also rumored to be developing its own A.I.-powered search engine, while the startup Perplexity AI has emerged as a significant player in the field. Valued at $9 billion, Perplexity AI’s A.I. search tools now manage approximately 20 million queries daily, a substantial increase since the beginning of 2024.

This surge of A.I.-driven search engines signals a crucial evolution in the tech landscape, with companies competing for market share by leveraging artificial intelligence to redefine how we seek and retrieve information online.