The impact of AI will leave no one or nothing untouched. With Cloudflare, we look ahead to the most important IT trends of 2025: AI is not surprisingly a recurring topic.
AI dominates all IT trends lists for 2025. And Cloudflare’s outlook for the new year is no exception. Cloudflare is optimistic about how AI will transform user experience and software development, but warns of the implications for computing, security and networking. We discuss what changes lie ahead with Christian Reilly, Cloudflare’s Field CTO for EMEA.
If you are not strong, you must be smart
Developing, training, and inferring AI requires the most powerful chips. AI companies like OpenAI and Meta are investing billions of dollars in infrastructure to keep their models running. According to Cloudflare, this can’t continue. The company is pushing for a paradigm shift in computing that’s not about brute force in the cloud, but about putting computing power as close to where it makes an impact as possible.
Reilly explains. “Training today’s (Gen)AI models requires high-performance and high-density computing . The kind of scale needed for this is often only possible in the cloud. Inference happens halfway between the network and the cloud. New AI applications that are sensitive to latency benefit more from inference as close to the end user as possible. A good example is self-driving cars, which need to be able to make decisions in an instant. Then you need smaller models that are tailor-made for specific applications and distributed computers that operate within that specific context.”
Double-edged sword
AI is not only making us rethink how we allocate computing power, it’s also changing security. Reilly calls AI a “double-edged sword.” “When introducing new technologies, there’s always a trade-off between the pros and cons,” says Reilly. “This is a general issue, not unique to AI. We already use AI for good purposes in security, to analyze behavior and patterns. But attackers are going to get smarter, too.”
AI is a double-edged sword for security. Defenders are getting smarter, but so are attackers.
Christian Reilly, Field CTO EMEA Cloudflare
One concern for Cloudflare is the potential spread of misinformation and how it can degrade the accuracy and integrity of AI models. Poor quality data literally turns AI into a “poisoned chalice,” Reilly warns. As users, we must not lose our critical spirit.
“We need to verify not only what goes into the models, in terms of the validity of the source data, but also what comes out. Verifying responses from AI systems is crucial to ensuring the quality and reliability of data. Sometimes it feels like you’re talking to a human, and humans make mistakes. With agentic AI, this evolution gets even more complicated, as systems now perform autonomous actions in other systems. Every company will need to draw a line for ‘acceptable use’: if the reliability of a system is too low, ask yourself whether you should be using it at all,” says Reilly.
Resilience
To be ready for the security challenges of tomorrow, Cloudflare says it is important for companies to build in sufficient resilience. Reilly refers to the dangers of vendor lock-in , when you are too dependent on a particular supplier. “The cliché says that you should spread your risk across three to five vendors.”
“Every organization has technology that improves the business. It is not always practical to consolidate on one vendor,” Reilly clarifies his point. “There is no single vendor that can offer everything. Every platform is different. Resilience helps you discover blind spots in your security and determine your core needs .”
Can you have too many vendors? “A security landscape that is too fragmented is of course not good either. It is not necessarily a matter of evolving towards more vendors, but towards vendors that have a broader scope. Security must be placed as close to the end user as possible,” Reily answers that question.
Consolidating on one vendor is not always a good thing, but too fragmented security is also not good. Security needs to be placed close to the end user and to the applications and data.
Christian Reilly, Field CTO EMEA Cloudflare
Cloudflare likes to apply the slogan ‘More is not always better’ to cybersecurity regulations as well. With NIS2 , Dora, the AI Act and other frameworks, the EU is fully committed to regulation to force companies to make their security a top priority. Cloudflare fears that, despite laudable intentions, the regulation misses its target and has no tangible impact on reducing cybersecurity.
We ask Reilly to clarify that position. “The challenge of any regulation is how quickly it adapts to new developments. Technology evolves quickly. The focus is on how to manage the risks of new technologies and in that context we will probably see more regulation. We should focus more on how to use technology for good. AI tools help to improve redundancy and increase the resilience of organizations.”
Same language
While we shouldn’t be blind to the potential risks, Reilly and Cloudflare are optimistic about how AI will transform our daily lives. “The big question is: what will be the impact on the human experience? Automation will change or replace jobs, but it will also create new jobs. I look at it pragmatically: AI has a positive impact on productivity. People are good at critical thinking and reasoning, which today’s LLMs can’t do yet ”.
“The way we work has changed a lot in the last decades. Technology has democratized work: Word processors, for example, once replaced the typewriter and democratized how letters were created,” says Reilly with a hint of nostalgia. “Some roles will inevitably become more automated. Every day we see new models that are better and further improve productivity. For that improved experience, you give up personal data. There is a need for frameworks that determine what data you are willing to give and in what context it is valuable.”
Reilly predicts a revolution in software development. “People are good at understanding processes and data flows in complex systems. Tools like Copilot offer unique opportunities to spend less time on coding itself and more time understanding processes and converting them into code. We don’t have to reinvent the wheel, but learn to understand how technology can improve and optimize productivity.”
“Be careful, to understand and debug code, you still have to learn it. There is no such thing as flawless code,” says Reilly, who disagrees with Jensen Huang’s assertion that coding is no longer a valuable skill. “The dynamics have changed,” he concludes. “In the past, we had to learn to speak the language of computers, now computers have learned to speak the language of humans.”
To understand code, you still have to learn it. There is no such thing as flawless code.
Christian Reilly, Field CTO EMEA Cloudflare
This editorial was created in collaboration with Cloudflare.