THE UK’S AI DREAMS MUST COMPUTE
Illustration created with Midjourney ©3LA.com
Prime Minister Rishi Sunak recently made a passionate speech claiming the UK could become a global leader in Artificial Intelligence. The idea filled patriotic hearts with pride and seemed plausible. After all, the UK has developed some seminal breakthrough businesses like DeepMind and ARM. However, it needs more processing infrastructure in order to support the demands of world class AI developers.
The GPUs simply aren’t here
GenAI requires massive amounts of compute - the computational horsepower to drive the LLMs on which it relies. The UK, is home to a fraction of the amount being consumed here. The rest, is largely provided by the US, via Transatlantic cables and satellites.
As of 2023, United States has the most supercomputers, with 150 machines. The United States has the highest aggregate computational power at 2,344 PetaFLOPS Rmax with Japan second (639 PFLOPS) and China third (455 PFLOPS).
As of 2023, United States has the most supercomputers and the highest aggregate computational power at 2,344 PetaFLOPS Rmax.
What are FLOPS and why Great British AI needs more
In 1964 the United States built the first supercomputer. It managed to process 3,000,000 FLOPS (Floating Point Calculations Per Second), or 3 megaFLOPS. In 1996, two US computers crossed the terascale barrier—that’s 1,000,000,000,000 or 1 teraFLOPS. The petascale era started in 2008 (1,000,000,000,000,000 or 1petaFLOPS).
In May 2022, the Frontier supercomputer at the Oak Ridge National Laboratory in Tennessee clocked 1.1 exaFLOPS, becoming the first exascale computer on record and (still) the current fastest supercomputer in the world. 1 exaFLOPS is 1,000,000,000,000,000,000 FLOPS, or as McKinsey recently explained it, 'an individual would have to perform one sum every second for 31,688,765,000 years to equal what an exascale computer can do in one single second’..
The 10 Biggest Computers In The World as of June 2023.
Exascale capability is critical to the UK fulfilling its AI superpower ambitions. Without it, it will be impossible to deliver world-class research and to serve as a hub for innovation. The UK has just 1.3% of global compute performance and its most powerful system ranks 28th. Exascale would offer new opportunities in research and development to grow the economy and enable developers to build more complex models and simulations. It would allow researchers to understand climate change, power the discovery of new drugs and advance the UK’s engineering capabilities. It is a critical component in empowering the UK’s leadership bid for AI.
Funding Hardware Isn’t Easy
While the UK is still ahead of its European neighbours in access to Venture Capital, it trails the US and China by a huge margin.
Venture Investments By Country source Dealroom.
UK Venture firms have leaned toward software investments because they largely require less capital and are often more responsive to market opportunities or legislative changes. While the UK government talks of giving academic computer scientists £900m to build a British supercomputer in Edinburgh, Amazon alone is spending around $25bn a year on compute. Amazon also intends to become an anchor investor in the upcoming public listing of ARM, the British chip designer. This would undoubtedly give its chip design and fabrication business some strategic advantages. Perhaps UK politicians should be incenting Amazon to provide more FLOPS and GPU clusters in Britain.
In other news, the German government has just offered Intel $11 (USD) billion to build two semiconductor plants in Germany. Intel will invest a further $22 billion. Taiwan Semiconductor Manufacturing Company (TSMC) is also going to build a €10 billion plant in Germany which will be 50% funded by state subsidy.
Why Does Location Matter?
Globalisation is under threat. Undersea cables and satellites are very vulnerable to the malevolent whims of bad actors. Recently China Telecom and China Mobile, withdrew their support of the Sea-Me-We 6 underwater network (19,200 km of cables connecting south-east Asia to western Europe). Though not directly related, this decision starts to question who builds, owns and manages the infrastructure underpinning the global internet. If the UK was deprived of its connections to the US, many critical systems would be effected. But UK AI would be gutted.
New Chips On The Block?
Nvidia’s dominance of the GenAI chip market has pushed up its share price by 240% over the last 12 months. TSMC, the company that fabricates Nvidia chips, can’t keep up with demand. Saudi Arabia recently bought 3000 H100 chips at $40,000 (USD) each and the UAE has secured similar numbers. “The UAE has made a decision to own and control its own computational power and talent” said someone familiar with the government’s thinking.
Escalating demand and diminishing availability is making prices skyrocket. Is there an opportunity here? It seems highly implausible for the UK to manufacture massively sophisticated chips any time soon. However, Nvidia’s near monopoly of the AI market might be challenged if British software companies could help to make AMD and Intel processors more appealing and competitive to AI developers wanting to train or deploy new models. An added benefit of this strategy succeeding would be to ease the worldwide shortages of Nvidia chips.
The Juice is Drying Up
Three west London boroughs - Hillingdon, Ealing and Hounslow - no longer have “sufficient electrical capacity for any new connections”. The west London electricity grid has been sucked dry by the explosion of data centres built there to connect to fibre optic cables that run along the M4 corridor, before crossing the Atlantic. Data centres gobble up vast quantities of electricity to power servers and ensure service are fault tolerant. Training AI models needs massive quantities of electricity. Without new sources of cheap, sustainable electricity, no organisation will consider locating Supercomputers or GPU Clusters in Britain.
What needs to be done
The UK government needs much more expertise to shape policy on compute. It needs to set a clear, manageable vision and a roadmap for the UK. All the interested parties - public and private - need to be brought together to accelerate the provision of world class resources. Even some basic frameworks to measure, monitor and forecast UK compute demands would be massively useful.
Public procurement also needs to be enhanced with the required technical expertise to shape decision making and investment strategies. Policy needs to consider and support the evolving demands of AI, quantum and other emerging technologies too.
Power generation policy has to become bolder and much more pro-active. The climate crisis is a major consideration but sustainable, carbon-free electricity is here already. Britain is second only to China in offshore wind farms. But to become an AI superpower the country will need the energy to power it.
