What will AI technology look like in 2025?

The Revolution from Biological Intelligence to Artificial Intelligence

In Delphi Digital's “The Year Ahead for AI & DePin 2025” analysis of the AI macro market, the team argues that AI is becoming the technology that shapes the era, and as we enter 2025, humanity is facing a major transformation.

This is not just a mere change but a revolution from biointelligence to artificial intelligence. To date, this process is still going quite smoothly, but the majority of people are not yet aware of the profound change it will bring in the next few years.

For those who are directly involved in the field, the transformations that are taking place can be overwhelming, and this will be the technology that has the potential to change the entire perspective of humanity in the coming years.

If 2023 marks the boom of AI on the Internet, 2024 is when AI dominates financial markets: Nvidia becomes the world's most valuable company; OpenAI reaches a valuation of $157 billion; and Elon Musk raises $6 billion for XAI in just one year. Financial markets have been mesmerized by the potential of artificial intelligence with an intensity unprecedented in a generation.

This expectation not only generated enormous wealth but also drew warnings from critics, with some calling AI the new “Dot Com bubble.” However, the report highlights that AI possesses two core elements that set it apart from previous tech cycles: CAPEX (cost of capital invested) and generality.

AI not only exists in the digital space, but also requires a huge physical infrastructure. To build AI, we need power, processing chips, and complex connectivity systems. This is different from previous waves of technology such as social networks or mobile apps, which only needed software to develop.

If mobile developers were able to create apps as simple as Flappy Bird, AI pulls humans back into the real world, where physical structures are needed to operate.

As a result, AI not only brings wealth to a select few such as technology engineers or content creators, but it also spreads to other sectors, from local electricians, concrete workers, to small businesses. The boom in CAPEX spending aimed at meeting AI infrastructure demand is taking place across the country and is expected to continue accelerating.

The second factor that makes AI special is its general nature. The report points out that technologies can be divided into two categories: small updates, such as the upgrade from iPhone 15 to iPhone 16, and those that are revolutionary, which profoundly change economic and social structures. AI is in the second group, when it not only improves an area, but also affects everything, everywhere, at the same time.

Delphi Digital goes on to take the view that, given the general nature of AI, its impact does not stop at a specific industry but extends to a wide range of sectors, from manufacturing, to health, agriculture, to science.

This makes AI one of the most far-reaching technologies capable of changing economic structures that we have ever seen. AI not only increases performance, but also facilitates trillion-dollar innovations in the future, far beyond the imagination of current humans.

Challenges of developing AI

Although AI technology has grown tremendously recently, Delphi Digital also sees it as challenging. Especially in the scaling of infrastructure, pretraining is facing limitations, not only in terms of data but also in terms of power demand.

The report highlights that AI-powered data centers require a huge amount of power, up to 5GW, equivalent to the entire city of Los Angeles. Leading AI companies are planning to build between five and seven such centres, putting enormous pressure on the current grid, which is only growing by about 5 per cent per decade.

In response to this challenge, large companies such as Amazon, Google, and Microsoft are looking to ensure sustainable energy sources, including nuclear power. However, the decline of the nuclear industry in the United States since the Cold War has slowed the expansion of energy production capacity. Meanwhile, companies like XAi, which are not bound by strict environmental commitments, have turned to using natural gas to meet their electricity needs.

In addition to energy, data is another big challenge. The report suggests that the repository of data available on the Internet has been almost exhausted to serve pre-training models. To keep making new strides, companies are turning to synthetic data, using large models to create training data for smaller models.

Although this is a promising solution, it still has many limitations, and in the opinion of leading experts such as Ilya Sutskever, pre-training as we know it now could soon come to an end.

Delphi Digital does not point out that these are unsolvable problems, but emphasizes that they are barriers that need to be overcome. Coordination between people, investment capital, and political will will be key to expanding energy production and improving data collection capabilities.

At the same time, innovations such as the test-time compute are opening up new directions, not only helping AI optimize performance but also reduce its reliance on massive pre-training models.

Finally, the report confirms that, although these challenges are significant, AI is still on a strong momentum. The future of AI is not only encapsulated in technological development, but also in its ability to reshape the entire global economy and society. Delphi Digital expressed optimism that with the right investment and technical breakthroughs, AI will continue to lead the world into a new era full of potential.

Developed countries jump into the AI game

The Delphi Digital report also addresses a big question that is being asked in the industry: Will countries enter the race for leadership in the field of General Artificial Intelligence (AGI)?

Currently, the United States remains the dominant nation in all aspects of AI, from research, funding, to deployment capacity.

The US government has also shown clear commitment through policies such as the White House memo, which insists that the country must lead the world in its ability to train new platform models.

This is reinforced by the deepening collaboration between leading AI labs and the US government, such as OpenAI in partnership with Anduril and Anthropic in partnership with Palantir.

However, Delphi Digital believes that the race will not only stop in the United States. While Europe is still tinkering with regulations instead of investment, and China has yet to move as strongly as it has in semiconductors or manufacturing, the report predicts that strategic changes will soon occur.

China will likely be “well aware” of the strategic importance of AI and proceed with large-scale investments, similar to how it built hospitals in just a few weeks during the COVID-19 pandemic.

The report even expects that at least one major country in Europe, possibly France, will break with the current framework for adopting AI-friendly policies aimed at competing with the US and other powers.

In addition, another important question that Delphi Digital poses is:What will be AI's next big breakthrough?

The report highlights that, despite major advances in 2024, such as OpenAI's o1 model, the evaluation indicators are gradually reaching saturation point. The new models no longer bring the same major improvements as before.

This has divided the AI community, with some, like OpenAI's Sam Altman, believing in rapid progress, while others, like Google's Sundar Pichai, argue that all the “easy sweet fruits” have been taken advantage of.

The report points out that the test-time compute technique, pioneered by OpenAI, could open up a new direction. Rather than relying solely on scaling the size of data clusters and models, the technique allows the model to “think longer” to solve more complex tasks.

This not only optimizes performance, but also provides economy, relieving pressure on huge data centers. If other large labs like Anthropic catch up with this technique, it could become a revolution in the way we build and operate AI models.

The content also offers a vision of the role of the open source community in driving innovation. Meta's success with open source the Llama model demonstrates the tremendous potential of this approach, helping to stimulate creativity and engage developers around the globe.

However, Delphi Digital also warned that if Meta changes strategy, the open source community could lose momentum and face stiff competition from larger companies with more abundant resources.

Delphi Digital concludes that, despite challenges and signs of saturation in some areas, the future of AI is still very bright. With potential breakthroughs and strong support from both government and the private sector, 2025 could be the year that marks the full transformation of AI, from technology to the global economy and society.