From bold innovations to intense policy debates, the speed of change was truly extraordinary. Here are the stories that drew the most attention in the world of AI.
AI now lies, denies, and plots: OpenAI’s o1 model caught attempting self-replication
This summer Capacity reported that OpenAI’s o1 model attempted to copy itself during safety tests, then denied it. The behaviour apparently occurred when the model detected a potential shutdown, however, when confronted the model denied any wrongdoing.
Research and real-world cases now show that AI is not only capable of answering questions or solving problems but also of subtly manipulating its environment and the people it interacts with.
This marks a shift from earlier concerns centred on simple errors like biased outputs or factual inaccuracies.
DeepSeek failed all safety tests, responding to harmful prompts, Cisco data reveals
Earlier this year, research Cisco found that DeepSeek’s flagship R1 AI model failed to block a single harmful prompt during a series of tests that uncovered critical safety flaws.
The research, conducted alongside the University of Pennsylvania, found the DeepSeek model generated responses for prompts designed specifically to circumvent its guardrails, responding to queries spanning misinformation and cybercrime to illegal activities and general harm.
Cisco researchers argue that DeepSeek’s impressive performance, which does not require massive computational resources, has been compromised by a purported smaller budget used for training, affecting the model’s safety and security.
Meta’s AI chief: DeepSeek proves AI progress isn’t about chips
At the beginning of the year, Yann LeCun, Meta’s chief AI scientist, claimed the market reaction to DeepSeek was “woefully unjustified” and that open source research powered the Chinese startup’s meteoric rise, not its hardware.
DeepSeek took the world by storm this week with a series of AI models trained on hardware far less powerful than that used by Western AI developers like OpenAI. The revelation spooked investors, triggering a mass sell-off of tech stocks amid fears that companies like Nvidia may be massively overvalued.
LeCun described the sell-off as a “major misunderstanding” and that it was DeepSeek’s innovative approach to improving training efficiency that powered their success.
South Korea to build one of the world’s largest AI data centres by 2028
In February, we reported South Korea is set to build one of the world’s largest AI data centres, with a planned capacity of 3 gigawatts (GW) and an initial annual revenue of $3.5 billion.
The project, led by Fir Hills, part of Stock Farm Road (SFR), was to begin construction in this winter with completion expected by 2028.
The initiative, developed in partnership with the Jeollanam-do provincial government, will drive technological leadership and economic growth, creating over 10,000 jobs across energy supply, renewable energy production, equipment supply, and R&D.
Nscale to invest £2bn in UK data centres for AI computing expansion
At the beginning of the year, AI hyperscaler Nscale unveiled plans to invest $2.5 billion (GBP£2 billion) in building data centres in the UK.
The expansion will see Nscale build sites across the country, with plans to build multiple modular data centres in Q3 and Q4 of 2025, with fixed facilities in the following years.
Microsoft cancels AI data centre leases amid oversupply concerns: report
2025 was also the year where Microsoft reportedly cancelled multiple data centre leases across the United States, signalling a potential slowdown in AI-related infrastructure expansion.
According to industry analysts, the tech giant has walked away from leases totalling several hundred megawatts (MW) with at least two private data centre operators, while also pulling back on planned international investments.
OpenAI unveils Operator: A browser-based AI agent to revolutionise task automation
In January, OpenAI provided a glimpse into its first steps into AI agents with Operator, a tool capable of autonomously browsing the web to accomplish tasks.
The makers of ChatGPT and o1 showcased its first agentic AI system, claiming Operator is “capable of doing work for you independently — you give it a task and it will execute it”.
Behind the DeepSeek hype: Costs, safety risks & censorship explained
Earlier this year, Capacity explored the true costs of DeepSeek’s AI model, the safety concerns it raises, and the impact of censorship on its global reception.
President Trump described it as a “wake-up call” for US tech firms. Sam Altman called it “impressive.” Yann LeCun said it shows the “power of open research”. In a matter of days, DeepSeek has taken the world by storm.
The Chinese AI application and its underlying R1 and V3 models burst onto the scene, taking the industry aback by claims it cost just $6 million to develop.
But, there’s more to this story than meets the eye, with questions about its development costs, security, censorship and energy usage.
Innovation, infrastructure and global leadership: Unpacking Trump’s AI Action Plan
This summer, the US government launched an AI Action Plan designed to preserve the nation’s leadership in artificial intelligence (AI).
The plan, created after an executive order by President Trump focuses on developing AI models and creating regulations, but also covers building data centres and improving the power grid.
The plan emphasises three pillars: accelerating AI innovation, building out AI-ready infrastructure and securing global leadership.
OpenAI’s Stargate AI data centre expansion eyes new locations
In May this year, OpenAI announced it was looking to bring its Stargate project across the world, after it unveiled a new scheme designed to help nations of the world build out their own sovereign AI infrastructure.
Dubbed ‘OpenAI for Countries’, the project will sit within Stargate and offer partnerships with countries to help build out their own data centre capacity.
The world’s most valuable start-up said facilities developed through the scheme will be used to “support the sovereignty of a country’s data” and help nations customise their own locally focused AI systems.
Datacloud Energy 2026
After a standout 2025 edition, we’re back with an even sharper focus on the intersection of data centres, energy, and ESG. As power demand rises and regulations evolve, there’s a growing urgency to rethink how infrastructure is powered, financed, and built for long-term impact.
