Computational complexity and power-hungry chatbots will be the unlikely drivers of electrification in the New Economy


When Battersea Power Station opened in 1933, The Observer described Sir Leonard Pearce’s masterwork as a “Cathedral of Power.” It wasn’t just the building’s grandeur that was impressive. At its peak in 1955, Battersea was generating over 500MW a week from four alternators and three coal-fired boilers, enough to keep the lights on for a fifth of London’s population.

 

 

Joshua Carson

Fast-forward 70 years, and our level of consumption is on a different scale. We’re using vastly more electricity – up to 30-times usage compared with the 1950s – whether that’s to power infrastructure, transport, industry, or to our homes. And how we source power has materially changed, ratcheting down our dependence on combustible fuel in favour of high-capacity renewables.

Though there may still be further innovation in energy efficiency, the demand on the grid will exponentially increase over time. Electric vehicles, the electrification of heavy machinery and the ubiquity of the internet are often cited as driving this trend, but there are others – like AI, quantum computing and cryptocurrency – that are deceptively power-hungry.

Consider Bitcoin. Today, a large crypto ‘mining farm’ can expect to eat 500MW a week due to the computational processing required to introduce new tokens into circulation. Each one requires a Battersea to run – and there are thousands of them. It’s astonishing to think that Bitcoin’s consumption could power a quarter of UK households for an entire year, equivalent to 0.13% of global electricity output (28.8TWh) every year, and this will balloon in line with computational complexity or ‘hashrate.’

To take another example, the sheer volume of ChatGPT queries submitted by its 100 million users is estimated to require 1GW every day. This is one application in a potentially infinite cosmos of large language models, (LLMs) and one with a lot of headroom if we consider that Facebook alone has three billion active users at any one time.

Computational demands on our electricity infrastructure will reach astronomical levels. I’d hazard to say that we’re in a constant state of underestimation. Even NVIDIA, the largest provider of AI hardware solutions, is projecting to ship 1.5 million AI server units – consuming 85.4TWh annually – by 2027.

And if Google adopted ChatGPT as its search engine, it would potentially need as much power as Ireland just to run.

Conservative estimates suggest that these computationally-intensive industries could consume 25% of all global energy produced by 2030. This is assuming that other revolutionary technologies, like virtual reality, aren’t by then invented, commercialised and expanded. To adapt, we’ll need new infrastructure that has the scalability to achieve and sustain large-scale power generation, while modernising legacy assets and building the supply chain that can suitably feed demand.

All this spells opportunity for investors in the so-called “new economy.” Especially in physical asset classes like real estate and infrastructure. Computational complexity and the computers that underpin these machine systems will need to be housed in bricks and mortar and supplied by a grid network that is fit for the future.

The most obvious requirement is for data centres and warehousing. Private equity titans and alternative investment managers like Blackstone are already partnering with data centres and cloud computing platforms to fulfil the “explosive growth of data.” In December 2023, Blackstone announced a $7bn joint venture with Digital Realty to develop four “hyperscale” data centre campuses across the US and Europe.

Even if we have enough data centres to manage and store computational information, these assets must be actively managed. Data centres have a “critical load” that is a function of storage capacity limited by cooling, power and space. Collectively, these limitations serve as a data centre’s “stranded capacity,” meaning that more data centres will be needed to ‘fill in’ the required bandwidth for those that are not reaching operating efficiency.

Investors with a different risk profile and a greater familiarity with infrastructure permit processes can instead look to the rising demand for renewable infrastructure. On a basic level, deeper computational strain on existing infrastructure and the imperative to decarbonise will push higher capex spend on renewable projects like the 3.6GW Dogger Bank in Scotland or battery-led power generation as in the case of gigafactories. These could provide much-needed additionality.

Then, there is the convoluted world of power allocation. While this will differ by geography and regulatory regime, data centres and renewable infrastructure share the common requirement of a grid connection to dispatch power or feed it in. For new projects, it is necessary to reserve a power allocation and take an option on its future use.

Regardless of how that is secured, either by title or by reservation, having some form of ‘right-to-connect’ with the grid makes ground-up development possible. It also unlocks the opportunistic potential of strategic land that is appropriate for the physical assets necessary for this new era of AI. Especially if policy is suitably flexible to fast-track large-scale and nationally-significant infrastructure.

Once the requisite physical land, grid, data and energy infrastructure is in place, the final ingredient is fibre connectivity. As with grid enhancement, we’ll need to lay fresh cables. It is here where there has been the greatest comparable obsolescence since broadband replaced dial-up, only for localised ethernet cable networks to be replaced with fibre optic. Fiera Capital’s fibre-optics investments, including Conterra’s 11,600-route mile fibre network across 21 states in the US, illustrate how independent bandwidth providers will be increasingly necessary to connect data centres in isolated locations.

The very fact that ChatGPT was unveiled less than two years ago and uses the same amount of energy per day as the entirety of new batter energy storage deployed in the UK last year, is stark evidence of how quickly computation-intensive industries are growing.

Governments competing for digital supremacy must make the path to capital deployment as simple as possible. Warm words to support the process of near-shoring industries like semi-conductors are not enough. This is without any doubt nationally-significant infrastructure. Policy must lead the case for supplying this supreme data-capacity challenge and expedite how quickly we develop its enabling infrastructure. Because if not now, then when?

 

 

Contact us


If you would like to find out more and see if we could help you, please get in touch.

Reputation Management.

Global Expertise.

Registered Office

2nd Floor, Berkeley Square House, Berkeley Square, Mayfair, London, W1J 6BD

Chelsea

109 Harbour Yard, Chelsea Harbour, London, SW10 0XD

Holborn

Central Court, 25 Southampton Buildings, London, WC2A 1AL

Phone: +44 (0)20 3514 0897

Reputation Management.

Global Expertise.