In the first eight months of 2024, major tech players including Microsoft (MSFT), Meta (META), Google (GOOGL), and Amazon (AMZN) collectively invested an astounding $125 billion in A.I.-related capital expenditures (CapEx) and operational costs, as detailed in a September report by JPMorgan. It is anticipated that the combined CapEx of these four tech giants will exceed $200 billion for the entirety of the year.
On the startup front, A.I. companies are witnessing record levels of funding from investors eager to tap into the technology’s lucrative potential. OpenAI is poised to conclude 2024 as the most financed A.I. company, boasting a valuation of $157 billion. Meanwhile, its competitor Anthropic is preparing for a fresh fundraising round that could see its valuation reach $40 billion.
With abundant funds at their disposal, leading A.I. companies must now demonstrate to both investors and the public that their significant investments in this emerging technology will yield returns. From the ongoing shift toward “agentic A.I.” to the exploration of new scaling laws and A.I.’s diverse capabilities, let’s delve into what 2025 holds for the realm of A.I.:
Agentic A.I. is set to be “the next giant breakthrough”
This term describes autonomous A.I. assistants capable of accomplishing tasks independently, without human intervention. The potential for A.I. agents to improve workplace efficiency and enhance daily living has caught the attention of many in Silicon Valley, with firms like Salesforce adopting agents as their next flagship product.
Microsoft has also made strides in this direction, unveiling a series of A.I. agents for its Microsoft 365 suite, including an agent capable of translating content into nine different languages.
OpenAI is on board with the “agentic A.I.” trend as well, with an upcoming model anticipated to handle tasks such as booking travel and writing code. Sam Altman, OpenAI’s CEO, remarked during a recent Reddit AMA that A.I. agents represent “the next giant breakthrough.”
The global market for A.I. agents is currently valued at over $5 billion, according to MarketsandMarkets. This figure could skyrocket to $47 billion by the decade’s end, largely driven by demand from enterprise clients.
Test-time compute may address A.I.’s data training crisis
A crucial factor in A.I.’s success has been the vast amounts of data fed into models. However, the available text, images, and videos on the internet are limited. To prevent stagnation in technological advancement, A.I. companies are exploring alternative training methods. One promising approach is test-time compute, where A.I. models enhance their performance by reasoning and taking longer to contemplate possible answers before responding. OpenAI’s o1 model recently showcased this theory.
During a November earnings call, Nvidia (NVDA) CEO Jensen Huang praised OpenAI’s new model as “one of the most exciting developments” in scaling, noting that “the longer it thinks, the better and higher-quality answer it produces.”
Huang’s enthusiasm is echoed by Microsoft CEO Satya Nadella, who highlighted test-time compute as a new scaling law in November. OpenAI co-founder Ilya Sutskever also mentioned it as an evolution in A.I.’s pre-training phase earlier this month.
Synthetic data presents a promising alternative
Another innovative solution to A.I.’s data shortage is the use of synthetic data—information generated by A.I. itself rather than traditional data sources. According to BCC Research, the synthetic data market is projected to explode to $2.1 billion by 2028, representing a staggering 450 percent increase from 2022.
Altman hinted at the potential of synthetic data over a year ago when he noted that “as long as you can get over the synthetic data event horizon, where the model is smart enough to generate quality synthetic data, it should be fine.” Companies like OpenAI, Anthropic, Meta, Microsoft, and Google have all begun incorporating synthetic data into their training and model refinement processes.
In October, the startup Writer unveiled a new A.I. model that was entirely trained using A.I.-generated data. This approach allowed them to significantly reduce costs, with development totaling only $700,000, a fraction of what other companies have spent. For comparison, OpenAI’s GPT-4 model reportedly cost over $100 million to train.
“Large world models” could revolutionize 3D A.I. environments
Much of A.I.’s visual output to date has been two-dimensional. However, tech innovators are aiming to change this in the coming years. “Large world models” are an emerging type of A.I. designed to construct interactive three-dimensional environments, transforming the realms of entertainment, gaming, and simulations.
World Labs, a startup founded by Stanford A.I. expert Fei-Fei Li, raised $230 million earlier this year to develop large world models featuring “spatial intelligence”—the ability to understand and interact with the real world. Li has used the example of a cat knocking over a glass of milk to illustrate this concept, emphasizing humans’ ability to predict and prevent such events.
In early December, Google DeepMind introduced its own large world model, Genie 2, aimed at simulating virtual environments for training and evaluating A.I. agents. This area is expected to be a key focus for the lab, as demonstrated by its recent hiring of Tim Brooks, a former OpenAI researcher who oversaw the video generator Sora. In a post on X, Google DeepMind CEO Demis Hassabis expressed excitement about “making the long-held dream of a world simulator a reality.”
A.I. search engines are set to transform online search
For years, Google has maintained an unassailable lead in the search engine market. However, as artificial intelligence progresses, a wave of AI-driven search engines is emerging to challenge Google’s supremacy.
Google itself is not standing still; in 2024, the company launched AI Overviews, a feature that provides users with AI-generated summaries instead of conventional search links. CEO Sundar Pichai is optimistic about the feature, predicting it will attract over 1 billion monthly users and improve overall search engagement and satisfaction.
Nonetheless, Google faces fierce competition from others in the search engine sector, including OpenAI and Microsoft, who are employing AI to enhance their services. Furthermore, Meta is reportedly developing its own AI-powered search engine, and the startup Perplexity AI has emerged as a significant player in this arena. With a valuation of $9 billion, Perplexity AI’s AI search tools now handle approximately 20 million queries daily, a remarkable increase since early 2024.
This surge in AI-enhanced search engines indicates a notable shift in the tech landscape, as companies compete for market share by leveraging artificial intelligence to redefine how we search for information online.
