It was way back in the mid-1950s when the phrase ‘artificial intelligence’ was first coined at an Ivy League university computer science summer workshop – or that at least is what my Google-generated AI overview reliably informs me.
However, as recently as 2018 few political folk had even begun to turn their minds to the awesome potential implications of AI on the world of work, yet alone its apparent questioning of what it means to exercise human will and free-thinking. Nevertheless, my curiosity was pricked when late one evening I came across an essay by Henry Kissinger with the provocative title ‘How the Enlightenment Ends’ which my private secretary had placed towards the bottom of my red box with the handwritten scrawl, ‘Minister, this might be of some interest…’
Anticipating that the then nonagenarian former US Secretary of State would wax lyrical on the national security and geopolitical aspects of AI, I was struck that his warning focused much more extensively on humankind being intellectually unprepared for the replacement of human cognitive processing by data-driven decision making. Almost a decade on l am still trying to figure out what this may mean to us all; will AI herald a new era of runaway economic growth and spark a fresh wave of scientific advance, or will it be the cause of extreme disruption to the global economy, democratic politics and represent an existential threat to humankind? As you may have already gathered some of these briefings are written as much for my own sake as for you, my hapless reader!
I suppose what worries me as we approach the twentieth anniversary of the financial crisis is that we appear to be on the cusp of repeating the same mistakes we made back then. Too much reliance on machine learning and the power of banks that were ‘too big to fail’ convinced us of the veracity and reliability of their financial modelling around risk. We became complacent in our unquestioning trust in data, innovation and technology. Is it not fair to conclude that we are in a similar place today as a small group of dominant Silicon Valley companies and their charismatic leaders usher us into a brave new world developed with self-interest at its heart?
The AI business model that has been created by global technology’s dominant players generates brand new content – from text to video, music and software code – by learning and adapting patterns from existing data. They have doubled down on this form of generative artificial intelligence by the wholesale adoption of Large Language Models (LLMs) which were initially trained on the entire internet to summarise documents and answer questions but are now evolving – through the creation of ‘agents’ that allow LLMs to perform tasks rather than simply generating text. As a consequence, technology hyperscalers have developed a conveniently self-serving narrative that global AI progress necessitates an industrial quantity of chips and tokens for the compute, colossal data centre infrastructure development and the instantaneous access to billions of parameters. This has led in recent years to what has been the largest capital expenditure boom in history – already at almost twice the level of tech infrastructure spending during the dot.com boom at the turn of the millennium.
As August dawned the financial pages were full of breathless coverage that Wall Street’s finest were underwriting a $500bn financing deal enabling Nvidia to become the lynchpin for the mobilisation of third-party capital into AI infrastructure supporting Nvidia’s wider ambitions. Few questioned the mounting interdependencies in ownership and cross-investment amongst the biggest players in the AI boom, yet alone the wisdom or ethical constraints that might apply to financing your own customers. It is salutary to note the extent to which chipmakers, hyperscalers, private-credit providers, data-centre owners and end-user large-scale, corporate customers are increasingly financing different parts of the same build-out. The business plans being relied upon by the next generation of investors may look very different when your erstwhile collaborators turn into fierce competitors.
Amidst all the hype it is understandable that many people conflate AI with LLMs – currently the best known are ChatGPT (Open AI) and Claude (Anthropic); what is less well understood is that there are plenty of other models, large and small, which can be trained on specific tasks without the requirement for the massive infrastructure expenditure on data centres. Indeed, some of these models and applications (the process whereby the end-user creates a bespoke service or system) can just as effectively and easily be trained and run on a desktop computer or hand-held mobile device.
There are perhaps two main reasons we hear so little about this competitive eco-system. First its most high-profile beneficiary to date has been the Chinese AI lab Deep Seek, whose R1 ‘chain of thought’ language model came to sensational prominence in January 2025. In contrast to comparable Silicon Valley products, R1 is far less expensive, much more accessible (China has promoted – for now at least — an open-source strategy in AI development) as well as free to use via the app and for developers to download and implement into their businesses. Other leading Chinese tech players from Moonshot AI to Alibaba have followed a similar path, launching low capital cost competitor products with comparably powerful capabilities.
Secondly, the methodology behind many Vertical Small Language Models (SLMs) being devised in research labs and universities across the world flies in the face of the approach that Silicon Valley wishes to impose upon us all. The ‘magnificent seven’ listed US tech giants, and their backers, are now engaged in an arms race with each of these fierce competitors betting the farm on the LLM approach which requires scaling up of existing product by adding more data and computing power to achieve peak performance. Hence the building of ever larger data centres with their colossal energy and water resource needs; it also explains the prevailing narrative of the most bullish Silicon Valley leaders that AI must bring with it unavoidable upheaval. To their critics the worldview of these self-styled Masters of the Universe is a toxic mix of utopian beliefs, delusional fantasies of omnipotence and snake oil salesmanship of the highest order. But the one thing the hyperscalers cannot possibly admit – especially with Anthropic and OpenAI having both indicated their intention to float within a matter of months – is that their business model is becoming commoditised by more modest, adaptable, cost effective, open source Chinese alternatives.
So, the future of AI is by no means preordained. If despite all the clamour for unprecedented further capex investment, the SLMs end up setting the template for the AI that we all rely upon in future, the current boom in technology investment may not lead to as severe a crash as many widely assume if a bust is to follow. For sure some of the more recent individual investors would lose their shirts, but the broader economic damage might be more restricted. Indeed, a fall in the retail price of AI might just as easily boost demand and become a saving grace even for many in the energy and chips supply chain apparently tied into the Silicon Valley way of doing things. Fears of unsustainable euphoria as the AI boom persists may be best tempered by drawing a parallel to the massive over-investment in railway infrastructure that came to a head in the 1860s and 1870s. The bursting of that bubble caused a lasting, deep slump in both the US and UK, but worthwhile facilities and networks had been created during the clamour to build infrastructure that contributed to economic growth in future decades.
On the other hand, some ominous signs abound. A rapidly and ever-increasing proportion of the financing of the AI boom has been secured on debt, especially within the already highly leveraged shadow banking sector, whose travails have caused more general concern. Whilst to date this has been little reflected in equity valuations, with the exception of some short-term volatility squalls, the credit markets are beginning to signal concerns. Our old friends, credit default swaps (namely insurance taken out in the event of bankruptcy) have registered record highs levels for leading semiconductor manufacturers such as Broadcom and Nvidia in recent weeks.
As I have already observed a red warning flag is rightly raised wherever new capital in a fast-growing sector is reliant on substantial investment from other companies already intertwined within its supply chain. Think back to the vast expansion in the derivatives market in the run-up to the global financial crisis of 2007 or the dot.com boom/bust at the turn of the century – as well as, for those with longer memories and a more parochial mindset, during the Lloyd’s of London reinsurance collapse of the 1980s.
Too many experienced City observers I speak to are deeply suspicious that with AI we are already deeply into bubble territory with vast amounts of US and European capital being overallocated; indeed, some would even suggest that resources are already being misallocated. What is clear is that an extraordinary amount of capital is being concentrated into one part of the AI eco-system, namely compute, chips, data centres and power infrastructure. Recent analysis from the Wall Street Journal suggests that the avowed capex figures materially understate the true extent of future commitments with the major technology giants already having off-balance sheet obligations over infrastructure leases and purchase commitments in excess of $3trillion.
In stark contrast with China, here in the West our headlong rush towards an AI future with its inherently uncertain road to monetisation is already crowding out other capital spending. By the same token the boom has to date focused on a small group of companies with hitherto strong balance sheets rather than an army of undercapitalised start-ups as we saw in the dot.com era. However, the next wave of largely private investment cash comes at a time of increasing hostility in the West to the entire business of wealth creation with populist voices being raised on equality and environmental grounds in opposition to the building of vast data centres; in the meantime, the general sentiment as to the pace of change likely to be felt in the workplace is one of bewilderment and foreboding.
By his own admission even Elon Musk veers from exhilaration to terror on a daily basis when thinking about the potential outcomes in the age of artificial intelligence to which he and his companies have contributed so much. On the other hand, the visionary Demis Hassabis who set up Deep Mind on these shores takes a less apocalyptic view; recognising that technological and societal progress is invariably a haphazard process, risky and often rushed. His optimism hinges on an appreciation that humanity invariably makes advances by muddling through via trial and error – although it is perhaps instructive that his own company has since 2014 been under the umbrella of Google, founded almost three decades ago and therefore a relative veteran of the US tech sector.
The sheer scale of AI’s existing capabilities yet alone its future impact on national and personal security, the world of work as we have experienced it and so much else has led to a clamour for its regulation. Leaving this task in the hands of its pioneers is generally regarded as too dangerous; their attempts to create either arms-length public interest safety boards or a non-profit structure are regarded as too self-serving to industry interests. The idealistic desire for a global protocol covering governance and oversight of AI may prove too daunting a challenge in these troubled geopolitical times. But these will be the themes of next month’s briefing.
Written on 8 September 2026 by The Rt Hon Mark Field, former Member of Parliament (MP) for Cities of London and Westminster and Consultant at Buchler Phillips, an independent boutique firm with an impeccable Mayfair London heritage, specialising in corporate recovery, turnaround, restructuring and insolvency.