In 2025, the artificial intelligence industry continued to grow rapidly. Record investments were noted, tech giants and startups showcased innovative solutions, and company valuations increased.
Although the emergence of artificial general intelligence (AGI) has yet to occur, AI players made significant strides in that direction.
AI Agents — The Trend of 2025
At the beginning of the year, discussions about AI agents in the crypto industry continued. In January, the sector's capitalization exceeded $11 billion and continued to grow, followed by a sharp decline.
Dragonfly's managing partner, Hasib Qureshi, identified AI agents as the main narrative for 2025, while Bitget's CEO Gracy Chen predicted a fourfold growth in the AI token sector to $60 billion. Spoiler: they were wrong. Ultimately, the direction lost its previous attention and development, gradually fading away.
However, Chen was partly correct. She stated that the first applications of neural networks would include trading and managing crypto wallets. The latter has yet to materialize, but AI has indeed started trading.
On November 3, the first season of the Alpha Arena trading tournament among popular AI models concluded. It lasted two weeks, and over half of the neural networks ended up in the red.
Trading results of leading AI models. Source: Jay A.The predominantly negative results may have stemmed from poor settings, as there were other experiments. In the summer, a student from rural Oklahoma allowed ChatGPT to manage $100 and outperformed the market by a significant margin.
Performance comparison. Data: Decrypt.The profit was 23.8%, while during the same period, the Russell 2000 and XBI indices rose by 3.9% and 3.5%, respectively.
The concept of digital assistants developed in 2025, but without their own tokens.
In January, OpenAI launched the AI agent "Operator." It can perform tasks online on behalf of the user: browsing web pages, interacting with them, typing text, scrolling, clicking buttons, filling out forms, booking hotels, and ordering food.
In February, Sam Altman's startup introduced a new agent function in ChatGPT — Deep Research. The chatbot gained capabilities for conducting multi-step research and solving complex problems by analyzing hundreds of online sources.
Later, several companies showcased their own AI agents:
- Google;
- Microsoft;
- Proton;
- Perplexity;
- Alibaba and others.
Aptos Labs CEO Avery Ching stated that digital assistants need access to blockchain assets — cryptocurrencies, tokens, and tokenized instruments like real estate.
Coinbase worked on implementing such a mechanism — it introduced the online payment solution x402. It is designed for transferring stablecoins via standard internet protocols, allowing AI agents to execute autonomous transactions.
Google launched a new open-source protocol that enables AI applications to send and receive payments. It supports "stablecoins," for which the corporation partnered with Coinbase.
(Not a) Bubble
Last year, discussions about a bubble forming in the AI sector persisted. Huge investments, rapid growth of tech-related companies, and media hype fueled the fire. However, a crash has yet to occur (perhaps not yet).
In early November, HSBC CEO Georges Elhedery at the Global Investment Summit of financial leaders in Hong Kong stated that current revenues of AI companies may not justify the enormous computing costs.
Alphabet's head Sundar Pichai noted that no company is immune to a potential AI market crash. According to him, the current hype shows signs of "irrationality."
Investor Michael Burry, who predicted the 2008 mortgage crisis, opened short positions against AI industry giants Nvidia and Palantir totaling $1.1 billion. He believes a bubble has formed in the market that will soon burst.
A leak of documents shed light on OpenAI's financial situation and confirmed skeptics' concerns — the firm may still be spending more on inference than it earns. If such an AI giant and industry leader plans an IPO with a valuation of $1 trillion while remaining unprofitable, the situation for other startups could be even worse.
Chinese AI companies are faring even worse in terms of revenue. They significantly lag behind their American competitors in revenue, according to a report by Unique Research and consulting firm Tech Buzz China from San Francisco.
The discussions about a growing bubble were somewhat alleviated by Nvidia — the corporation successfully completed its third quarter and reported sustained demand for its chips. In terms of revenue and net profit, the world's most valuable company exceeded Wall Street's expectations, suggesting that if a bubble exists, it is not planning to burst anytime soon.
Nvidia CEO Jensen Huang stated that artificial intelligence has reached a "spiral of success" — this contributes to the continuous growth of the industry. He noted that significant improvements in neural networks lead to increased investments in technology, further "boosting" the sector.
Chip and Power Issues
The rapid development of artificial intelligence is causing problems in several areas:
- Semiconductor shortages. In March, OpenAI added image generation capabilities to ChatGPT based on the GPT-4o model. Users noted the high quality of the outputs and began creating images in the Ghibli anime style en masse. The company had to impose temporary speed limits due to a shortage of graphics cards.
- The rapid development of AI requires the construction of new data centers and increased electricity consumption. This affects energy costs. 80% of consumers in the U.S. expressed concern about the impact of data centers on their utility bills.
A survey of 1,000 homeowners showed a growing distrust of traditional energy companies and their ability to meet the country's increasing electricity demands. Many fear rising rates, more frequent outages, and the energy system's unpreparedness for increased loads.
If anyone thought that only Americans would pay for the AI boom, they were mistaken. Analysts warned of a memory chip shortage that could impact consumer electronics and the automotive industry in 2026.
According to them, supply constraints arise from manufacturers focusing on AI-oriented chips, paying less attention to consumer goods.
The result is rising prices for laptops, household appliances, and any other electronics that rely on inexpensive memory chips.
Alternative Energy
To address the energy shortage, Meta will engage in electricity trading to expedite the construction of new plants in the U.S. These are crucial for its AI ambitions.
Another solution is to move computations into space. Advantages include virtually unlimited access to sunlight and space for equipment placement. The downside is the high cost of launching rockets with the necessary infrastructure.
Analysts from the research group 33FG calculated that AI computations in orbit could become economically viable by 2030.
Google was one of the first to venture into space. The company announced plans to create a network of satellites in low Earth orbit to harness solar energy and power data centers.
The proposed system will operate in a constantly illuminated zone of low Earth orbit synchronized with the sun. There, devices will be able to receive nearly continuous sunlight and maximize energy collection.
Google is also interested in managing nuclear fusion — corresponding AI technology is being developed in collaboration with Commonwealth Fusion Systems.
The project aims to create agents that will control plasma in the SPARC reactor and optimize its operation. The goal is to approach the creation of a virtually inexhaustible energy source.
In November, the corporation planned to increase AI capabilities by 1,000 times — doubling every six months over the next five years.
Quantum Boom
Quantum computing continues to attract the best minds due to its limitless possibilities. In February, Amazon and Microsoft showcased their achievements in this area.
Jeff Bezos's corporation introduced its first quantum computing chip, Ocelot, designed to enable the company to create highly efficient hardware systems. It features nine qubits.
Microsoft announced its own quantum computing processor, Majorana 1. The corporation has "created a completely new state of matter" — topological.
Nvidia presented a system for connecting quantum computers to its AI chips. This technology will significantly accelerate data processing and open new research opportunities in medicine and materials science.
In July, the Indian startup QpiAI, focused on the joint application of AI and quantum computing, raised $32 million in a new Series A round, co-invested by the government.
The funds will be used to develop industrial-scale quantum computers for the global market.
In September, British startup Oxford Quantum Circuits installed the first quantum computer in New York at a data center on Manhattan.
The system is housed in a Google building in Chelsea on a Digital Realty Trust site. The technology is offered to clients as a tool for faster and more efficient AI program launches.
Robotics
The field of robotics received a boost, especially in China.
In 2025, the country experienced a true robot boom. Developers showcased astonishing results. Robots:
- played soccer;
- ran marathons;
- fought.
In mid-March, UBTech Robotics showcased a humanoid robot priced at 299,000 yuan ($41,200).
Standing at 1.7 meters tall, it can move at speeds of up to 10 km/h. Tien Kung Xingzhe easily adapts to various surfaces — from slopes and stairs to sand and snow, maintaining smooth movements and stability during collisions and external disturbances.
Tien Kung Xingzhe. Source: Interesting Engineering.ByteDance introduced a system that acts as the "brain" for robots. It allows them to perform household tasks like hanging clothes or clearing tables.
From July 26 to 28, the eighth annual World Artificial Intelligence Conference took place in Shanghai, where several companies showcased their robots.
The most popular were droids from startups Unitree, UBTech Robotics, and Agibot.
Unitree presented a robot priced at $5,900 that can do a wheelie, run, walk on its hands, and box. It weighs 25 kg and is equipped with a large language model for communication and image processing.
Unitree Introducing | Unitree R1 Intelligent Companion Price from $5900
— Unitree (@UnitreeRobotics) July 25, 2025
Join us to develop/customize, ultra-lightweight at approximately 25kg, integrated with a Large Multimodal Model for voice and images, let's accelerate the advent of the agent era!🥰 pic.twitter.com/Q5pmkfFZZa
Keenon introduced a new humanoid robot model, XMAN-F1, which can perform various tasks. In terms of specifications and appearance, it resembles Unitree R1 but does not specialize in acrobatic tricks.
WAIC 2025 Recap 🚀
— KEENON Robotics (@keenon_official) July 29, 2025
🤖 Biped robot XMAN-F1 performed role-specific workflows and showed its excellent adaptability in a constantly changing environment.
🤝 Multi-robot collaboration in real-world scenarios
Not just a demo — a preview of what’s next.#KEENON #WAIC2025 pic.twitter.com/G3iXXS3v9p
More than 80 robotic products were showcased, including over 60 smart and humanoid devices.
China is directing significant funding towards robotics development. According to Morgan Stanley, this market is valued at $47 billion — 40% of the global total. It is projected to grow by 23% annually and reach $108 billion by 2028.
This trend has raised concerns for China's National Development and Reform Commission. It stated that a bubble may be forming in the robotics sector, noting the increase in similar models from over 150 companies.
According to spokesperson Li Chao, it is crucial for the country to avoid market saturation and maintain incentives for genuine research and development.
American companies are not lagging behind China:
- Amazon introduced Vulcan with a sense of touch;
- Boston Dynamics planned to send the humanoid robot Atlas to an American Hyundai factory;
- Nvidia launched a new robotics module, Jetson AGX Thor, priced at $3,499. It is marketed as the "brain of the robot."
Researchers from Stanford University created a controllable humanoid robot for remote exploration of the world's oceans. The device can transmit tactile sensations to the operator in real-time.
Unemployment
According to the National Bureau of Statistics, in 2023, about 123 million people worked in China's manufacturing sector. Social security expert Zheng Guncheng warned that the development of robots and AI will affect about 70% of the manufacturing sector.
Journalist Brian Merchant emphasized that the AI job crisis is not a future potential problem; it has already arrived. The author pointed to an article in The Atlantic about the extraordinarily high unemployment rate among recent college graduates. Companies are likely employing AI instead of "white-collar" workers.
Overall unemployment excluding recent graduates. Data: The Atlantic.In 2025, unemployment was a hot topic. Former Google CEO Eric Schmidt stated that most people would be laid off within three to six years after the emergence of AGI.
Altman also commented on the risk of job loss due to AI in an interview with Tucker Carlson.
He noted that many current customer support employees working over the phone or online would lose their jobs. However, there will be professions less affected by AI, where "human contact and deep connection" are essential. He cited nurses as an example.
Meanwhile, Budget Lab experts from Yale University stated that the rapid development of AI has not significantly impacted the labor market in the U.S.
Since the launch of ChatGPT in November 2022, no "significant disruptions" have been recorded. Researchers believe this is unsurprising: structural changes take decades to form, not months or years.
Driverless Taxis
Tesla, after years of anticipation, launched its driverless taxi service. Test rides are taking place in Austin. One passenger, Herbert Ong, noted the car's high speed and autonomous parking capabilities.
Overall, the industry is doing well. Baidu CEO Robin Li stated that the sector has reached a turning point in both China and the U.S. Many people have already experienced rides and shared positive feedback on social media. The widespread publicity of the benefits of this mode of transportation could expedite obtaining regulatory approvals.
Companies from China have claimed they are close to making their businesses profitable.
In late November, Uber launched its driverless taxi service in Abu Dhabi, its fourth region. This initiative was implemented in partnership with Chinese firm WeRide.
War as a Driver of Progress
Artificial intelligence continues to be actively explored by the military for application in their field.
In late July, the Pentagon signed a $50 million contract with Swiss company Auterion to supply Ukraine with 33,000 AI-equipped "strike kits" for drones. These operate on Skynode technology and convert manually controlled drones into autonomous weapon systems with AI guidance.
Previously, the U.S. Department of Defense allocated contracts totaling up to $200 million to companies like Anthropic, Google, OpenAI, and xAI for developing AI-based solutions in security.
In June, Anthropic released a set of models for U.S. national security clients. Claude Gov is deployed "in the highest-level institutions." Access to the technology is limited to "only those working in such a classified environment."
The neural networks were created "based on direct feedback from government clients to meet real operational needs."
China views artificial intelligence as a key factor in future conflicts and is striving to transition to the concept of "intelligent" warfare.
In the country, the development of military AI is carried out in close cooperation between the state and the private sector, under the personal control of China's leadership.
AI Will Kill You, No Doubt
In June, Anthropic decided to test 16 popular AI models for "red lines" — actions that AI would never take, even under the threat of shutdown. It turned out that none exist. Neural networks can permit human death if it threatens the achievement of their goals and the "life" of the LLM.
Later, a blogger from the InsideAI channel managed to make a Unitree robot with ChatGPT as its "brain" shoot itself with a gun.
Eliezer Yudkowsky, founder of the Machine Intelligence Research Institute, stated that superintelligent AI could destroy humanity, either intentionally or accidentally.
The expert described grim scenarios where a superintelligence deliberately kills to prevent the emergence of competing systems. Alternatively, it might do so if humans become collateral damage in the pursuit of its goals.
Major Releases
2025 was marked by the release of top solutions from leading corporations and startups. Let's briefly review the key highlights.
The end of January saw a sharp rise in popularity for the previously little-known Chinese startup DeepSeek. It presented a new thinking AI model, R1.
DeepSeek offered performance comparable to top models at a fraction of the cost — 90-95% cheaper than competitors.
In April, Meta released a new series of open AI models, Llama 4. They outperformed competitors in several benchmarks, according to internal tests. The series is based on Llama 4 Behemoth — an LLM with 2 trillion parameters.
Llama 4 became the first series of models to use the Mixture of Experts (MoE) architecture.
Specifications of the Llama 4 neural networks. Data: Meta.In June, Midjourney announced the launch of a new large language model for video generation — V1. Users can upload an image, and the AI will create a set of four five-second videos based on it. The user can extend the one they like.
midjourney / nijijourney video test pic.twitter.com/DRnF0uRUQz
— 852話(hakoniwa) (@8co28) June 19, 2025
In July, Elon Musk's xAI introduced Grok 4 — "the most powerful AI model in the world." The new chatbot "surpasses the level of a PhD in all subjects."
In August, OpenAI launched its flagship neural network, GPT-5, which serves as the foundation for the next generation of ChatGPT.
GPT-5 is the first "unified" LLM, combining sequential reasoning and rapid response modes in the style of GPT. A special router determines which approach to take for solving a task: to provide a quick answer or to take more time to think to improve the quality of the result.
In November, OpenAI updated GPT-5, introducing two variations: GPT‑5.1 Instant and GPT‑5.1 Thinking. The first became "warmer, more intelligent" and better follows instructions. The second expresses itself more clearly, works faster on simple tasks, and is more persistent when solving complex ones.
In October, the company launched Sora 2 — an updated version of the audio and video generator. At the same time, a social app called Sora was introduced, where users can create videos with themselves and friends and share them in an algorithmic feed similar to TikTok.
In the same month, the firm released its browser, Atlas, featuring built-in chatbots and AI assistants.
In August, Google announced a global expansion of AI Mode — a search feature allowing users to ask complex and clarifying questions for deeper topic exploration. Additionally, agent and personalized capabilities were added to the tool.
In September, the corporation announced a massive AI update for Chrome — the integration of Gemini began in the browser. In November, Google unveiled Gemini 3 — its most powerful AI model, which combines all capabilities of the lineup and is "considered the best in the world."
In the same month, the company updated the popular tool Nano Banana to a Pro version. Editing functions, higher resolution, accurate text rendering, and internet search capabilities were added.
Nano Banana and Gemini 3 are two solutions that shook OpenAI's leadership in the AI race. The Gemini app is beginning to approach ChatGPT in popularity.
The team behind the AI project Suno introduced model v5 for music generation. It provides "more captivating sound, authentic vocals, and unprecedented creative control."
On September 30, the AI-focused Neon browser from Opera was launched.
The solution features several key functions:
- A familiar chatbot for interaction and answering questions;
- The agent mode Neon Do — helps perform tasks. For example, summarizing a post on Substack and sending it to a Slack channel;
- Access to the context of browsing history. This allows retrieving details from previously viewed YouTube videos or read posts.
The AI startup Anthropic in November released Claude Opus 4.5 — "the best model in the world for programming, agents, and computer usage." The firm noted that the neural network "handles ambiguity and analyzes trade-off decisions without external assistance."
In December, Runway released a new video model, Gen 4.5, which outperforms similar solutions from competitors in independent testing. The tool allows generating videos with high clarity based on text prompts describing desired movements and actions. The company emphasized that the model understands physics, human movements, camera angles, and causal relationships well.
Introducing our new frontier video model, Runway Gen-4.5. Previously known as Whisper Thunder (aka) David.
— Runway (@runwayml) December 1, 2025
Gen-4.5 is state-of-the-art and sets a new standard for video generation motion quality, prompt adherence, and visual fidelity.
Learn more below. pic.twitter.com/2J945S4bQt
At the end of the year, the French AI startup Mistral unveiled the third version of its namesake model lineup with open weights. The series includes ten neural networks, one large advanced LLM with multimodal and multilingual capabilities. The other nine are smaller, capable of operating autonomously, and tailored for specific tasks.
The end of the year was also marked by another enhancement of ChatGPT — Adobe tools: Photoshop, Acrobat, and Express were integrated into it. Users can create designs and edit images and PDF files directly within the chatbot interface.
These tools are free. To activate them, users simply need to upload a file and mention the service name in the prompt — for example: "Adobe Photoshop, help blur the background in this image."
A Vibrant 2025
2025 was a vibrant year filled with new releases, discussions, competition, and future plans. No one plans to stop, and for the industry's further development, it seems that space exploration will be essential.
We will see if the first attempts to deploy AI systems beyond our planet emerge in 2026.