Transition in a bubble environment

Having been in New York earlier this week for our REFF North America conference where all things renewable energy were discussed, it’s curious that the classification of gas-fired power plants as “transition” raised not the glimmer of a knowing smile.

The term is long established and “natural gas” – nobody says methane any more, like it stinks, or something – has long been classed a “transition fuel”… primarily in emerging markets.

In the EM case, its transition status was granted to denote the replacement of coal-fired thermal plants [shudder], diesel-fired [gasp] and oil-fed pollution belching monsters [retch].

Sitting in New York’s glorious Met Club on Monday/Tuesday, gas was repeatedly touted as a transition fuel to feed the monstrously large data centres that we’re told will make this world a better place.

Transition to what? Renewables? Nuclear? In the US, might that even be coal?

Or is this just a massive fig leaf? A worldwide slap in the face to Net Zero ambitions as the infra community wheels out this established power generating technology, but justify it by referring to it as a stop gap measure to fuel a rapacious beast.

Sadly, gas is the only viable option to meet challenging data centre deadlines… while knocking new-world thinking into a cocked hat… with airy mentions of carbon capture and storage… in the full knowledge CCS / CCUS cleansing option will be left on the table in the vast majority of cases.

As one infra grandee puts it: “The argument for gas as a transition fuel held water when it was about keeping the lights on as we switched off coal. This isn’t that. It’s just an excuse to keep doing what we’re doing.

“Net Zero is dead. We tried. We failed. The story really should be about mitigation.”

Energy mitigation

While it’s painful to give McKinsey credit, in April it published research that shows by the end of this decade – 2030 – data centres worldwide will require $6.7 trillion of investment to keep pace with the demand for compute power.

According to the McKinsey report: “Data centres equipped to handle AI processing loads are projected to require $5.2 trillion in capital expenditures, while those powering traditional IT applications are projected to require $1.5 trillion in capital expenditures. Overall, that’s nearly $7 trillion in capital outlays needed by 2030.”

And to power investments in data centres, it’s estimated this will require 300-500GW of new generation… which, at the lower end of that scale, is the equivalent of 100x Hinkley C nuclear power plants.

Well now. Small modular reactors were mentioned once or twice at REFF. Heck, IJ did a podcast with Blue Energy the other day on SMRs that’s worth a listen. And while this technology will be part of the solution, gas will be a lot faster and easier.

Maybe it’s time to step away from the transition fig leaf and just call it what it is.

As one infra banker of many years standing says: “I love the concept of ‘transition fuel’. Supply chain issues, political instability in once very stable countries, immense costs and dwindling public support are slowing down momentum on energy transition. Banks are changing their ‘no more oil and gas’ policies… and that says it all!”

AI bubble burst

Talking to folk around the industry this week and there’s consensus that it ain’t gonna end well and quite a few shirts are going to be lost.

Some put it quite simply when looking at the business model, asking: “How much do you pay for ChatGPT?” Meanwhile others have more a micro perspective: “There’s a reason that data centre leases have break clauses.”

Further, sources reckon this community ain’t held in the highest regard: “The tech people view the infra people as quite stupid, but with cheap capital.”

While everyone agrees that AI is coming like an express train, they also see the fuel tanker – frantically waving driver standing perilously close to vehicle – broken down on the tracks ahead.

One infra banker says: “There are no obvious revenues at the moment for any of the many AI applications and I assume a lot of development costs. As AI is ‘everything’ on steroids, regulators and companies will need to consider how to protect data.

“But – like the internet, the cloud and so on – AI will be sustainable once revenue streams appear and security is strong. The sheer size of applications and potential to users is the obvious revenue stream, but it will take some time and many will get their fingers burnt.

“It will take time but then it will be upon us… like HAL 9000 in 2001 Space Odyssey… but 35 years late.”

A contact of the infra fund persuasion points to “consumers rebelling about energy prices leading to restrictions on data centres or differential pricing”, adding that “open AI” will be more useful to end users – therefore less data and energy intensive – than “closed AI”.

This source helpfully clarifies that “closed AI” is the US model where the AI does everything – takes your query, searches the internet and delivers the answer. Meanwhile, “open AI” is where someone sells users the source code so they can create AI of their own to mash their own data.

But what’s going to cause the bubble to burst? Most agree that will be greed and FOMO (fear of missing out).

One infra specialist with a tech bent says: “Those who understand this stuff are either starting to stand back and say WTF or doubling-down and hoping they can outpace the tidal-wave of reality coming through.

“This is specifically the LLM (large language model) scaling problem and the reality that LLM does not get us to AGI (artificial general intelligence) – a plateau that has been known about for some time.”

This source reckons that some investor-types are caught in the classic “losing sight of the ground” dilemma.

Said source says: “I see comparisons here to what preceded the LMX (London market excess of loss) spiral of reinsurance in the mid-1990s, the tech-bubble a few years later, construction-contractors chasing turnover for share price (less concerned if it’s decent work), and the CDO packaging of junk pre-the GFC.

“In all cases people starting buying to sell or Ponzi returns derived from the next round of chumps.

“Classic FOMO mentality where understanding of what lurks within is nil and opportunity to flip is on overdrive. Then suddenly the music stops. When all stops, this pops... again!”

Meanwhile and infra fund guru adds: “No one believes the price of the Big 7 [Alphabet (Google), Amazon, Apple, Meta (Facebook), Microsoft, Nvidia, and Tesla] is the present value of their future profits… but they do know that calling the top too early is nearly as bad for a fund manager as being caught in the crash.

“If everyone gets burned – especially all the trackers – you might beat the index and be upper quartile of the actives… even if you lose a third of your client’s money… while everyone else loses half.

“Whereas if you bail out now, you might get sacked for underperformance over the next 3 quarters before the crash actually happens.”

So, let’s jump on the express train with a tinge of dread, knowing what lies ahead.