As the new year begins, Andriy Silchuk, DataArt’s Head of R&D Center and Delivery Director, looks back on a turbulent 2025. From defining trends and high-profile scandals to breakthrough innovations and rare bright spots, he recaps what shaped the IT and hi-tech world—and shares his outlook for 2026.
Ladies and gentlemen, we’re lucky once again to have made it to the end of the year, so let’s officially tally up the year-end results.
As before, let’s follow a familiar route: we’ll briefly recall last year's forecasts, then look at the major trends, scandals, the good, and the bad that we all experienced in 2025. We’ll also examine the big names we lost this year, determine the heroes and villains, and recall 2025’s surprises. Then we’ll end the program with our 2026 forecasts.
So, pour yourself your favorite drink – and let's go!
A Brief Look at Last Year’s Forecasts
At the end of 2024, we predicted that we’d get real AI agents, turbulence in the US IT industry, changing requirements for engineers, "data as oil," and a growing lag between Europe and the United States in 2025.
What we actually got:
- AI agents have truly made their way from being the subjects of presentations into full production: ChatGPT agents and a bunch of other tools are already walking around the sites themselves, pressing buttons, executing scripts, and the Linux Foundation is even launching an initiative based on agentic AI standards.
- Turbulence in the United States IT industry hasn’t gone anywhere: antitrust lawsuits against Google/Meta, content wars, regulation — it was all present again this year.
- The requirements for engineers have changed drastically: "I know how to work with AI tools" is now a basic required skill. Big Tech is introducing KPIs for using AI, and those who resist are told to look for a new job.
- Data and infrastructure are truly the "new oil." The problem is not even about data, but about servers, GPUs, memory, and energy — everything is becoming more expensive and scarce.
- Europe has cemented itself as the regulatory champion, with the EU AI Act, record DSA fines for X, etc. In the US, meanwhile, they’re more so debating how to regulate IT, rather than actually doing any regulating.
Our predictions were right practically across the board. Not because we’re prophets, but because the trends were blatantly obvious.
Four Most popular trends of 2025 (they’re all about AI)
1. Agency AI: From "chats" to real assistants
2025 was a turning point: the focus shifted from "generative AI" to agentic AI. ChatGPT agents and similar systems no longer just respond, but also perform tasks themselves—they open websites, monitor statuses, book, write, and edit documents. Businesses are churning out their own agents for support, sales, and back office, DevOps, and domain tasks, and at the same time, a whole zoo of multi-agent frameworks is growing. We've officially gone from "a chat who advises something" to "an assistant who does the job but still needs supervision."
2. GEMINI, GPT, CLAUDE and others – a new level of "smartness"
Google has finally shown that it’s still alive and very strong, with its Gemini 2.x and 3 models, Nano Banana, and other tools, and deep integration into Search, Android, and Workspace. OpenAI rolled GPT-5/5.1 out of "thinking mode" and made it the default in ChatGPT, effectively dragging a bunch of niche tools under it. Anthropic with Claude 4.5 is seriously putting the pressure on its competitors in coding and reasoning. Meta continues to pump Llama in open source. For the user, it is no longer "one model is better than another", but a whole forest of ecosystems that fight to be your main "superstructure in work." The high level of competition is always in our favor.
3. Data centers as new "capitols" with Heroes 3
The capitol is an extremely necessary thing in Heroes III, but it costs a lot of money. It’s the same with data centers. The IEA and the European Commission predict that data centers already consume about 1.5% of all electricity on the planet, and could double this consumption by 2030, largely due to AI. Energy demand in the United States for data centers jumped 20% year over year, and AI servers are taking an increasing share of capacity. Big Tech is responding in its own style: it’s buying up solar/wind plants, it’s building gas and small nuclear projects, it’s turning old coal-fired power plants into data centers, and its signing contracts for building its own nuclear power plants. Nuclear power plants, Karl!
4. Regulations, courts, and "AI psychosis"
The EU AI Act has officially started, and DSA is starting to bite with real fines. In the United States, state attorneys general issue warnings to large AI companies about mental health risks. The first lawsuits have appeared where ChatGPT and other models appear in real tragedies — from suicides to murders, where AI allegedly added fuel to the paranoia. Regulation has traditionally lagged behind technology, but politicians and courts are already in play, and 2025 has clearly shown that "it's just a chat" no longer works as an excuse.
Five most high-profile scandals of 2025
1. DeepSeek: China's "nightmare" for the market
At the beginning of 2025, DeepSeek released its models with an embarrassingly low price and pretentious claims about "ridiculous training costs." The market panicked, NVIDIA shares sagged. Then it turns out that everything is not so simple as "cheap" training. The quality of "supermodels" from China took a hit, too. But the shock of just how one release can collapse half the market remains.
2. X becomes DSA’s first major "patient"
The EU decided to demonstratively apply the Digital Services Act and issued X an estimated $140 million fine for manipulative blue ticking and refusal to provide data for research. This is the first major case DSA in action, and hardly the last. The signal is clear: playing "I do whatever I want" in Europe won’t work anymore, even if you really love freedom of speech in your own interpretation.
3. TikTok: Banned, then not banned, with an eternal "window for agreement"
The TikTok saga in the US was reminiscent of a soap opera. The law required that either TikTok sell itself to an American owner or leave the market. TikTok defiantly shut down its service in the United States before the deadline, the administration dragged out the time, and then Trump came. He extends the "window for agreements" several times (he’s the master of the “art of the deal,” let's not forget). As a result, bidding continues for a year, names of possible buyers are announced, but they never do get full control — formally everything has been signed, but in reality everyone just pretends to be very busy, and postpones the final steps.
4. Google: Antitrust wars and the shadow of selling Chrome
Google is simultaneously focused on several different fronts: dominance in advertising and search, abuse of its mobile platform, and the use of the web to train models. Against this background, there was even a lot of talk and rumors that the company could be forced to sell Chrome, and there was a long queue of those who wanted to buy it. Hyenas can sense blood from far away, as they say. The sale didn’t take place, but the very fact of discussing the sale of the #1 browser shows how tightly Google was squeezed.
5. OpenAI's exit from Microsoft's influence
OpenAI and Microsoft are officially "restarting" their partnership: Microsoft remains a large shareholder, but without total control. Azure's exclusivity is being diluted, and some OpenAI services are moving to other clouds. In the end, the companies declare friendship, but in fact they are preparing for a "civilized" departure: OpenAI wants to make decisions on its own and have freedom, while Microsoft wants the right to develop its own AI separately. At least, that's what they say. OpenAI gets certain advantages from separating, that's clear, but what Microsoft will get out of separation is still a question.
Three most positive events of 2025
1. AI in Medicine: From promises to real-world treatments
2025 was the year when AI in medicine finally showed something more serious and applicable than just promises on paper. Results of clinical trials of AI-developed drugs against cardiovascular and oncological diseases are emerging, and Rentosertib for the treatment of idiopathic pulmonary fibrosis has demonstrated safety and benefits. Furthermore, AI approaches to early diagnosis of cancer and liver diseases are actively developing. This is not yet "AI cured cancer," but real steps in this direction are already being taken.
2. Quantum Computers: Less hype, more benefit
After years of promises about how cool quantum technology can be, we are slowly but surely moving towards practical applications. New systems like Quantinuum, Helios, and Google Willow show progress in bug correction and stability. It's still expensive and niche, but it looks less and less like PR, and more and more like a long-term bet.
3. Global IT demand comes to life
India's IT services exports grew by about 12.5% to $224 billion in fiscal year 2024-25 after several sluggish years. For the industry, this means a simple thing: enterprise money is again used not only for cost optimization, but also for new projects and digitalization. For Ukrainian outsourcing, this means not a direct contract, but a very positive indicator: customers are ready to buy again. If the money returns to India, then it will reach us.
Four worst IT news in 2025
1. Massive declines in global services
In October, there was a long-term crash in AWS us-east-1, which in turn “crashed" Slack, Atlassian, Snapchat, and a million more. In November and December there were two big Cloudflare failures, one of which knocked out up to 28% of the world's HTTP traffic. The conclusion is banal, but painful: the Internet is too dependent on several infrastructure players. The words "multi-region/multi-cloud" on presentation slides are not a guarantee of real sustainability.
2. AI-enhanced cyberattacks
Cybercriminals are awake too: tools like PromptLock are emerging, which use generative AI to automate phishing and more complex attacks. The year 2025 saw a series of major leaks and ransomware attacks on energy, logistics, and other critical systems. AI increases the productivity not only of developers, but also of all the "bad guys".
3. Giant data breaks
Prosper Marketplace in the United States lost the data of 17.6 million people, and the South Korean company Coupang lost another 33.7 million accounts. In total, there are more than 50 million records with names, addresses, documents, and order histories. The reputation of fintech or e-commerce can now be lost in one bad year.
4. Mass layoffs in tech
According to TrueUp and other trackers, in 2025, almost 700 waves of layoffs in tech companies took the jobs of more than 200,000 people — an average of 600+ dismissals every day. The headlines are the same again: Amazon, Microsoft, Google, Intel, Meta, etc all laying off employees. Increasingly, companies are saying bluntly: we are cutting people to invest in AI and automation. So either we learn to work with AI, or AI will replace us, little by little. Don't forget this simple rule.
Six most interesting releases and announcements of 2025
1. ChatGPT Atlas and Comet — AI-browsers
OpenAI launched ChatGPT Atlas — a Chromium browser with ChatGPT at its heart: sidebar, summarizing articles, comparing products, working with documentation directly in a browser window. Perplexity rolled out Comet — also on Chromium, but with a focus on a personal agent who does its own research, deletes unnecessary tabs, and rakes mail. These are no longer add-ons on top of Chrome, but a new class of products: "a browser as a shell for an AI agent."
2. AGENTS.md — README for agents
In August, AGENTS.md appeared — a simple file at the root of the repository that explains to AI agents how to live in a project. How to collect and test code, where the entry points are, and what the rules are. In just a few months, tens of thousands of repositories pick it up, GitHub adds guides, and the Linux Foundation with OpenAI/Anthropic formalizes it as part of the standard for agentic AI. Starting this year, documentation is divided into human-made (README.md) and agent-made (AGENTS.md) — and it looks like it’s here to stay.
3. Claude 4.5 is a "programming neighbor" for developers
Anthropic updated its entire lineup: Opus 4.5, Sonnet 4.5, Haiku 4.5. Opus seriously improves reasoning, long contexts, and tool/agent handling. Sonnet has become a workhorse at an adequate price. Haiku has become an ultra-fast, high-volume option. In reviews, Claude 4.5 is often cited as one of the best dev assistants for real-world projects, not just for template tasks or pet projects.
4. Gemini — Google shows it can do it all again
Google rolled out Gemini 2.0 (Flash / Flash-Lite), then 2.5 Pro / Flash / Deep Think, and at the end of the year, Gemini 3 Pro. The models are getting faster, smarter, and are heavily tied to the Google ecosystem. The most important thing is total integration: Gemini lives in the search, Gmail, Docs, Android, and Google AI Studio. This is no longer an attempt to catch up with competitors, but a separate ecosystem that can really be used on its own. Many note that this is one of the best AI releases of the year.
5. Starlink Direct-to-Cell and Ukraine
SpaceX launches commercial Starlink Direct-to-Cell: satellites work as base stations, SMS texts are sent from ordinary smartphones through space without special devices! And then Kyivstar becomes the first operator in Europe to launch D2C together with Starlink: first for SMS and basic messages, then they plan to add voice and mobile Internet. For Ukraine, this is not just another feature, but an important element of resilience during blackouts and shelling.
6. Bonus: Sora and the first step to a "dead internet"
OpenAI released Sora 2 for video generation and a separate application — a conditional "Instagram," purely for AI videos, called Sora. Feeds are clogged with synthetic video, people are delighted, and at the same time, many are wondering: if social networks begin to massively switch to generated content, how much "live" Internet will we have left? On my side, I can admit I myself sometimes get caught up in this content. And yes, sometimes I can't even distinguish it from real content.
Five "most" interesting hardware inventions of 2025 — once again it’s all about the metal
1. Most innovative device: the Meta Ray-Ban Display
The first AR computer to be really similar to a daily device, in the form of normal glasses. Meta AI's messages, navigation, translations, and replies are all right in sight. Special attention should be paid to the Neural Band, a bracelet that reads muscle impulses and allows you to control the interface with gestures. There are still plenty of questions about the product, but as far as a direction of innovation goes, it’s very interesting.
2. Hobby of the Year – Logitech MX Master 4
Yes, it's "just a mouse," but the MX Master 4 with the new Haptic Sense Panel and Actions Ring was one of the most pleasant changes to the working day. Ergonomics, multi-device, and a bunch of custom shortcuts that really save time. As the owner of the previous version, I can honestly say: this is a device that’s difficult to pass up.
3. Disappointment of the Year – iPhone 17 Pro
On paper, there’s the A19 Pro, a new camera, Apple Intelligence, and marketing bull shit. In practice, there’s a controversial design, aluminum instead of titanium, and the main AI features arrived late to Europe and in a stripped-down form. If you’re looking for an Apple, then check out either the iPhone 17 Air, albeit with questions, but at least it’s something new, or the regular iPhone 17, which turned out to be much more successful.
4. An Interesting Niche Product – Oura Ring 4
Although Oura Ring 4 was released last year, it received a cool ceramic version this year, and looks like the king of niche devices: tracking sleep, stress, and activity in the format of a beautiful piece of jewelry, not just another screen on your arm. Not for everyone, but for those who bother, wellness is a very nice gadget.
5. Garbage of the Year – Samsung Galaxy XR
Formally, it’s the first Android XR device and flagship in cooperation with Samsung, Google and Qualcomm. In fact, it’s a rather expensive demo. Albeit lighter than Vision Pro, it’s not very ergonomic, with an external battery, damp software (very raw), unstable tracking, and a poor catalog of applications. Against this background, even the already-mentioned, not very successful Vision Pro looks cooler.
Big IT names we lost in 2025
The traditional block where you want to press F and cry.
- Bill Atkinson was a legendary Apple engineer, creator of MacPaint and HyperCard, and the man who shaped the look of early GUI.
- Steve Shirley is a pioneer of outsourcing and remote work, the founder of Freelance Programmers, who built an outsourcing business long before it became mainstream.
- Margaret Boden is one of the founders of cognitive science and AI research, and the author of classic works on the interaction of artificial and human intelligence.
- David Benaron is a doctor and entrepreneur whose developments formed the basis for the sensors of modern fitness trackers and smartwatches.
- Udo Kier is an actor, but for us he is forever Yuri from Command & Conquer: Red Alert 2.
This is only a small cross-section of the people whose work "lies quietly under the hood" of the things we use every day, and that we have lost this year.
And separately — R.I.P. Skype, a piece of our everyday life, to which time still said "that’s it" and left.
Other 2025 Highlights
IT Hero of 2025 — Jensen Huang, CEO of NVIDIA
Under his leadership, NVIDIA briefly touches the $5 trillion capitalization bar on October 29, becoming the most valuable company in the world. The demand for their chips is rewriting records, and NVIDIA itself has finally turned from a "company for gamers" into a monopolist of infrastructure for generative AI. The man in the black leather jacket became the face of the era – more than anyone else.
2025 IT Villain — Astronomer CEO Andy Byron
We could easily give the statuette to one famous billionaire again, but this year the anti-hero award goes to Astronomer CEO Andy Byron. He became famous not for his products, but for his very loud personal story and the memes around it. Sometimes the villain of the year is not the one who breaks the market, but the one who coolly spoils his reputation because of an affair at a Coldplay concert. The story will go away, but the memes will stay with us forever.
IT anecdote of 2025
On the one hand, there’s Ilya Sutskever and Mira Murati, who collected billions for a startup based on a "bare name," without a product. It's very cool, but I would believe in such a joke only in an anecdote.
On the other hand, there’s a wave of madness around the new image generation model in GPT-4o: the Internet is turning into an anime carnival, Sam Altman complains that there’s not enough power, and users can’t stop. True surrealism.
IT Surprise of 2025 — Oracle and Media Triples
Oracle suddenly becomes an AI cloud star: its shares soar more than 40% in a day after news of giant contracts and OpenAI connections, its capitalization approaches a trillion, and Larry Ellison overtakes Musk in the ranking of the richest people in the world by several hours.
In parallel, Netflix, Paramount, and Warner Bros. Discovery play out a complex love triangle with purchase claims and political overtones. The content market is shrinking, and we are gradually approaching the world of "one app for all videos." Jobs willing, one day it will be so.
Mobile 2025: Liquid glass and Epic vs Apple
This year I decided to add such a nomination. Apple is importing a complete redesign of iOS in the style of liquid glass - beautiful, loud, uncomfortable in places, but definitely hype.
And Epic is finally winning a small but important victory in the fight against Apple. It was definitely a pebble that, albeit a little, changed the issue of commissions in mobile stores. Not a revolution, but it is from such microcracks that large monopolies begin to gradually rethink their behavior.
Five predictions for 2026
Alright, let's move on to the forecasts!
1. AI agents will become the new daily software, and the hype will continue
In 2026, the average engineer will have not one chat or tool, but several AI agents who will do the routine: walk through Jira/Confluence, rake mail, and write drafts. The item "experience in building and managing AI agents" will increasingly appear in vacancies.
2. Energy will be the main limitation on the AI boom
We’ll see the first cases when the construction of data centers is directly limited by access to energy and water. Investments in energy, especially nuclear energy, will become a part of Big Tech's AI strategy.
3. Regulators will move from chaotic fines to a system of rules
The first real AI certification frameworks for medicine, finance, and education will appear in 2026. They will still be bureaucratic, but they won’t look like chaotic steps any longer. At the same time, we can expect high-profile court cases against AI platforms for damage to health, people’s wallets, or their reputations.
4. Fake AI profiles and content will become commonplace
What now looks like "strange Insta accounts" and individual cases will become a massive buzz in 2026. Generated faces, stories, news, bloggers, individual content will become the new normal. The question of the year will be "is there anything real here?"
5. Internal AI platforms will become the standard for companies
If in 2025 proprietary LLMs or internal AI platforms were a feature of a few, then in 2026 an internal AI platform with access to documents, code, and processes will become a new "corporate standard." Someone will buy ready-made solutions, someone will assemble it themselves, but "enterprise AI" will cease to be a pilot, and will become an obligatory part of the infrastructure.
Closing 2025
This year was difficult. At times it was extremely difficult. For many people it became the most difficult year in their entire career and life. But from the point of view of IT, this year turned out to be incredibly rich. AI became smarter, data centers became hungrier, regulators got angrier, Big Tech got fatter, and Ukrainian IT got even more inventive.
*The article was initially published on DataArt Team blog.
Top comments (0)