What a rollercoaster ride AI is proving to be. Its machine-speed DNA has already turned the assimilation of raw data and accumulated human knowledge into information available almost instantly to everyone, commoditising knowledge, if not necessarily judgement. In doing so, AI is collapsing innovation time horizons, rewriting productivity assumptions and disrupting almost every sector. Not to ignore the impact it is having on our very human cognition as I wrote about earlier – The Executive Mindset Challenge in the Age of Machine Speed Intelligence.
If that was not enough, the story is now taking another fascinating turn. AI is no longer simply transforming technology and business models; the extraordinary capital required to feed its appetite for compute is beginning to stimulate the creation of an entirely new class of financial instrument, one whose underlying asset may become obsolete almost as quickly as it can be financed.
I am referring to the announcement by Nvidia on 10th August’26 that it has signed memorandums of understanding with six of the world’s largest financial institutions to create compute financing platforms designed to mobilize over $500 billion in third-party capital for the construction of AI infrastructure globally, a deal that redefines the chipmaker’s role from hardware supplier to financial intermediary.
The emerging model for financing AI infrastructure raises an uncomfortable possibility, could AI succeed spectacularly while many of the investors funding its expansion still lose money? I suspect there is a real risk that could be an outcome …
The logic behind the current boom is straightforward. Demand for AI compute is growing rapidly, Nvidia remains dominant in advanced Graphice Processor Units (GPUs) and financiers increasingly see chips and contracted compute capacity as infrastructure assets capable of supporting large pools of debt. The risk is that the economics of AI may change faster than the financing structures built around it.
Today’s investment assumptions depend on continuing demand for enormous quantities of compute. However, AI models are also becoming more efficient, hardware is improving rapidly and alternatives such as custom silicon are advancing. If useful AI becomes dramatically cheaper to produce, demand must grow even faster simply to keep all the new infrastructure economically productive. That creates a potentially dangerous mismatch, where technology may evolve on an 18 to 24 month cycle, while financing may assume assets retain meaningful value for 5 years or more.
The warning sign will therefore not be Nvidia’s revenues. It will be GPU utilisation, rental pricing and residual asset values. If capacity continues expanding while utilisation and compute prices fall, financing may have started creating supply faster than genuine economic demand can absorb it.
History offers plenty of precedents. Railways transformed economies, yet many railway investors lost fortunes. The internet changed the world, while telecom and fibre investors went bankrupt and the lessons of cloud that have largely been ignored by organisations are coming home to roost as AI adoption is being impacted by technical and cyber debt.
AI may follow exactly the same pattern. The technology can be revolutionary without every financial structure built around it being sound. Perhaps Wall Street’s newest innovation is not financing AI but finding a way to securitise technological obsolescence before anyone has had time to depreciate it.

Leave a Reply