It hardly needs to be said that AI has become the dominant force in technology this year, seizing the attention and investment of businesses, individuals, and government entities alike. For telecoms, like many other industries, that is unlikely to change going into next year. Juniper Research recently published its ‘Top 10 Telco Trends 2024’ with AI making an appearance in three.
But this list, and indeed, much of the conversation around AI and telecoms, is missing something. There’s so much focus on what AI can enable, from network intelligence to voice bots, but few seem to recognize how important telecom networks are going to be for AI to live up to its huge hype. NVIDIA and their GPU chips have been the star of the show so far when it comes to enabling infrastructure, but you cannot cash in on your NVIDIA chips without connectivity.
Forecasts from IDC earlier in the year predicted global spending on AI-centric systems would reach $154 billion in 2023, with this due to almost double to $300 billion by 2026. If this happens, then we are currently only scratching the surface of AI’s explosive growth. The UK government is adamant about being a major player in the space and is two years into a ten-year national strategy to make the UK a ‘global AI superpower.’ Again, however, there is no mention of ensuring telecom networks are ready to deliver on all of this promise.
Gartner predicts that by 2025, generative AI will account for 10% of all data produced worldwide. It is less than 1% today. As more businesses start to leverage and build AI applications and generative AI tools like ChatGPT and DALL-E, the amount of data traveling through networks back to data centers will explode. IDC says generative AI alone will create zettabytes of technology in the next five years.
Telecom networks will need to carry and backhaul all this extra traffic back to data centers, this is before even considering the continued growth of other data-hungry technologies like 5G or the IoT.
To support this, building a dynamic computational infrastructure will be key, supporting AI from the network edge to data centers.
According to research from 650 Group, almost 1 in 5 Ethernet switch ports that data centers purchase will be related to AI/ML and accelerated computing by 2027. Adequate Points of Presence and strategically placed data centers will be essential to manage this rapid information flow in a cost-effective and sustainable manner. Some hyperscalers have already recognized this, with Microsoft announcing plans to invest £2.5 billion to build next-generation AI data centers in the UK.
The placement of these data centers will also be crucial to maintaining the high speeds and low latencies that modern networks and real-time applications demand. This will require further data center investment around London and further afield. For the former, however, the increasing cost of land and lack of access to adequate power could halt investment by an alleged £500 million.
Investment in new facilities is already moving North to cache data near hubs like Liverpool and Manchester, but expanding AI data centers in the Nordics are expected to generate substantial data traffic to the North West of the UK, prompting further investment in data centers in Northern England as AI evolves. If London’s data center investment becomes too untenable, the balance of power (or data, to be precise) could shift.
However, to ensure AI works effectively across the UK, it’s not enough to just build data centers. The ability of the networks to transport large amounts of data back and forth to these data centers for processing will also be key. This means having a network infrastructure that is not only widespread but also able to handle the heavy data traffic necessary for AI applications. Additionally, internet service providers will need to store data close to the user so that services can be provided quickly and efficiently, tailored to their specific location.
As large data centers transition to faster, more scalable infrastructures, high-capacity connectivity will be essential to keep pace with the ever-expanding number of users, devices, and applications. The OEM ecosystem has done a great job at pushing the boundaries of what is achievable by pushing the laws of physics, and 400G proliferation continues at pace, but the industry will increasingly embed 800G and even 1.6TB solutions.
So, despite all the exciting and potentially transformative applications of AI for telecom networks, the two have a far more codependent relationship than many realize. If the nation’s fixed infrastructure is not equipped to manage the increasing volume of data, its AI ambitions could flounder. Telecom providers must then prepare to support the growing demands of AI. Businesses investing in AI will also need to consider network partners capable of meeting their needs. Likewise, the government needs to recognize and consider supporting the vital role these networks will play in shaping the UK’s AI future.
