Blain’s Morning Porridge October 2^nd 2025 – What if Hyperscale AI is Just a Gold Rush?

 “If I could bottle electric and sell in cans, I’d be a billionaire…”

What if we are completely wrong about AI? A revolution in how tech is being innovated across the economy is clearly happening –  but what if we’ve got the winners and losers wrong? What if we’ve misjudged how AI will evolve as companies and individuals innovate AI solutions for themselves, seeking to preserve their data sovereignty and control of  their systems?

The headlines this morning are all about OpenAI’s $500 bln valuation, making it the world’s most valuable start-up. A whole new raft of millionaires has been created as the staff sold stock. It may prove excellent timing. Bloomberg describes hyperscale AI as an investment frenzy. Never has a market felt so bubblicious. Never has the concept that currently unprofitable firms are about to be worth billions (by spending trillions through hyperscale AI infrastructure spending) felt so…. utterly frothy. It has the feel of a gold-rush.

Having read Jack London, I have my doubts – but anyone who regularly reads the Porridge will know I tend to be optimistically sceptical.

My first point would be that economic success is about energy.

If there is one clear lesson, it’s invest in power generation. If AI is a Klondyke Market, then don’t dig for gold, but sell the miners what they need – electricity. Power prices appear to be going one way:

• The massive demand for electricity to power and cool data centres serving the AI expansion is driving electric prices higher. That demand for data centres has triggered massive new building of more data centres – which will push energy costs even higher.
• An unbreakable rule of economic growth is rising energy demand pushing up costs is a prime driver of inflation. Electric inflation (“lekyflation”) will push up interest rates and drive down profits.
• Higher energy costs are likely to be a key factor in holding back broad growth in the west. Energy will flow to the sectors that can afford it, further fuelling the rising concentration risk in the Mega 7 (or is it 10) AI linked stocks, inflating the bubble!
• And, when AI pops… there will still be plenty of broad demand for the picks and shovels of economic growth – electricity!

My second issue with AI is… What if we’ve got it all wrong….?

What if it’s all a false feedback loop? What if OpenAI and the other hyperscalers prove to be the “Betamax” of the age. (I’ve attached a wiki-link to explain the Betamax paradox – that anything better doesn’t sell if the alternative is cheaper and almost as good – to anyone under 40.)

I am not the only person asking.

One of my chums, Michael Power, ex-Ninety One, has written as fascinating article in the South African Daily Maverick – Where does AI go from here? He has been thinking about the future of AI as infrastructure, rather than products. He heretically suggests consumers may decide innovating their own AI tools – rather than using big models – is a better and more efficient use of their own data.

The American way of AI is the American Way. The AI giants have pioneered closed-weight proprietary systems, which demand the massive infrastructure build out and loads of power for the all-encompassing LLMs that can crunch all the world’s data into the system to spew forth the most likely reply – without revealing the parameters behind it (hence closed-weight).

Their LLMs require the most expensive chips and computing, massive data centres, dedicated electrical power sources, and as much water as possible to cool the plant (and if water is scarce, then use more electric to drive the coolers.) They include ChatGPT, Google’s Gemini, and Anthropic’s Claude. The prospect of hyperscale purchasing has made the pick and shovels infrastructure even more valuable – Nvidia for chips, Oracle in data-centres, and Amazon and Microsoft in the cloud.

Closed-weight systems are energy dependent. They require burning as much fuel as required to drive the turbines to produce barrel after barrel of electric, to feed the machine. After unravelling President Biden’s efforts to build-out renewable power in the USA, Trump plumped for the “drill, baby, drill” mantra. Just this week, he opened the door to renewed coal mining, critically aware of the rising demand for power is not being matched by new energy plant construction.

The closed-weight AI models, like Open-AI’s GPT, have driven the AI narrative to its current heights. They market themselves as the only way forward on every advertising window. They encourage users to innovate AI solutions  using their LLM to churn out answers. They don’t allow firms to use their proprietary models to run programmes on the users own systems, or change the internal “weights”. It’s great for the firms that own the models.

But… it’s less good if you care about transparency and data sovereignty.

In contrast, China is using its copious supplies of hard-won renewable power to build a very different approach to economic growth. They aren’t denying climate change or the need to defend the environment. But they are absolutely pro-power building windmills and fossil fuel together – they understand the future. (They will use fossil fuel to drive the current economy, and use renewable power to create the future zero-carbon solutions  – like e-SAF; “power-2-liquid (PtL) sustainable aviation fuel made through new carbon capture technologies and green hydrogen.)

(Greta Thunberg and the environment have rather gone out of fashion for the time being – but it will likely become a top priority again when warming oceans smash some typhoons into major conurbations, and the anti-woke right wing struggles to blame ESG and the lefties for a deeper crisis. The point is: the USA has closed the door to renewables, China is investing.)

Partly because of export-controls on chips, but also because they understand the economics of energy, China is pioneering open-weight AI systems using cheaper (available) chips and less power to build smaller language models for users to run on their own software to analyse their own data sets – critically allowing users to retain sovereignty over their data!

It’s not just China – many open-source western AI firms like Meta, and Mistral in France, see the future offer similar options that allow users to develop their own approaches to innovating AI in their businesses. The word on the street hints there is a fundamental shift from closed to open-weight systems underway. I am told top Silicon Valley investor A16Z (Andreessen-Horowitz) is seeing AI start-ups pitching off Chinese models, and Alibaba’s open Qwen3 tops the rankings.

The gist of Michael Power’s argument is that open-weight AI is ultimately cheaper, if not free! It’s more efficient, more accessible and more adaptable than the big closed-weight systems. I would add it’s also less likely to be exploited by monopoly-seeking US closed-weight firms monetising our own data against us.

Let me quote from a note Michael sent me:

“Wall Street still bets on the incumbent triumvirate: Closed-Weight LLMs, Nvidia chips and hyperscale clouds. But history warns us: when Microsoft dominated the world of the 1990s with Windows and Office, it thought the game was over and won. Then came Open-Source Linux… which begat Open-Source Android. Today, Android powers 45% of all operating systems worldwide. Microsoft has 27%, mostly on laptops and desktops. Apple has 21%. To its credit, post-2000 Microsoft has survived by adapting its business areas into completely new areas… including Closed Source AI and cloud computing.

Today’s AI titans — Alphabet, Meta, Oracle, NVIDIA, and yes Microsoft — are valued on the belief that proprietary models are the only path to dominance. But what if the next trillion-dollar AI company isn’t built in Silicon Valley? What if it’s built in Hangzhou – where Alibaba gives away Qwen so freely, even rural clinics in Bolivia can use it?  What if it’s built in Hyderabad – where a startup fine-tunes a 7-billion-parameter model on local dialects, serving millions who speak Tamil, Telugu or Bengali?”

This kind of alternative thinking further persuades me the market’s current love of closed-weight AI is a bubble.

SLM’s open-weight systems require a fraction of the power of LLMs. Even cloud computing may become optional – say’s Power.  At the moment Microsoft and Google are using first-mover advantage to favour their position in closed-weight AI. They have the data which open-weight systems lack. There are now hybrid systems which combine open-weight transparency with closed-weight monetisation opportunities.

The point is.. don’t assume the AI multi-trillion datacentre investment world is the only way forward. Think about air-travel – we are at the stage where the AI closed-weight Airships are the height of luxury and are moving passengers from A-B efficiently in comfort. Aircraft are string and fabric constructions that are anything but comfortable…. But.. Concorde is just 30 years down the road…

What is that hissing sound…? Is it the hydrogen leaking out the current AI bubble..? Time to move on to the next fantabulous thing..!

And aint it funny OpenAI is a closed-weight firm…. ? Says it all…

Out of time and back to the day job….

Bill Blain

CEO – Windshift Capital

Author – The Morning Porridge

Partner – Shard Capital