Amazon Web Services (AWS) has defended the environmental and economic rationale behind its push into custom silicon and AI infrastructure, amid increasing scrutiny over the carbon footprint and power demands of hyperscale cloud providers.
Speaking at the AWS summit in London, Dave Brown, vice president of compute and networking at AWS, highlighted the firm’s investment in bespoke processors like Graviton and Trainium, which he claims offer significant improvements in price-performance and energy efficiency.
“Graviton gives you more performance at a lower price, and uses 60 per cent less power than its alternatives”, Brown claimed when speaking to City AM. “The best type of power is the power you didn’t have to use at all”.
AWS says its chips, built using ARM architecture – a type of processor based on a simpler set of instructions for operations -, were designed with efficiency in mind, and are part of Amazon’s broader decarbonisation efforts.
The leading tech giant has committed to powering operations with 100 per cent renewable energy, by 2025.
UK in focus
The London region has emerged as a key node in AWS’s global infrastructure.
Brown confirmed that AWS’s total investment in the UK has now hit £6.3bn, with further capacity being added through data centres and local zones.
“We’ve been very happy with the London data centre since 2016”, he told City AM, citing broad UK uptake from start-ups to major clients like Natwest and Nando’s.
AWS recently added two new local zones within the UK and is pledging capacity expansions to maintain what Brown called “the illusion of infinite scale”.
Yet, industry analysts have questioned whether such grown is sustainable given the mounting pressure on national grids.
Research by the International Energy Agency warns that AI and data centres could double global electricity consumption by 2026, while a UK government-commissioned report this year urged faster planning reforms to support energy infrastructure.
Energy constraints and competition heat up
AWS claims to be mitigating energy constraints by placing data centres in areas with surplus green energy, connecting to users via ‘local zones’.
But with AI training workloads ballooning, there is escalating concern about whether custom silicon alone can offset surging demand.
“Trainium 3 offers around 50 per cent power reduction over Trainium 2”, Brown claimed. “That allows us to do more with the same amount of power”.
While AWS pitches chip diversity as a customer advantage supporting the likes of Nvidia and Intel alongside its own silicon, analysts have noted it is also a hedge against supply chain risk and geopolitical shifts.
Brown didn’t wish to comment on these global tensions, such as US-China tariffs, but acknowledged them, claiming that AWS are closely monitoring advancements.
