SoftBank’s sale of its Nvidia stake and holdings in two telcos says something about the shifting nature of investment flows within the artificial intelligence industry. It also says something about how the insatiable demand for capital to fund the industry might test the limits of financial markets’ capacity to supply it.
Japan’s SoftBank cashed out its $US5.8 billion ($8.9 billion) holding in Nvidia and has raised more than $US11 billion from the sale of shares in T-Mobile and Deutsche Telekom to help fund a $US30 billion investment in OpenAI and its share of the $US500 billion Stargate data centres joint venture with OpenAI and Oracle.
Masayoshi Son’s Softbank has more than $US300 billion of assets, but its commitments to OpenAI and Stargate and a $US1 trillion AI and robotics manufacturing hub in Arizona are stressing its financial capacity and liquidity.Credit: Bloomberg
It wasn’t a loss of faith in Nvidia, the $US4.7 trillion chipmaker at the centre of the AI boom, that prompted the sales, but the need to raise the funding for the financial commitments SoftBank’s Masayoshi Son, who is nicknamed “The Warren Buffett of Japan”, has made to OpenAI and the Stargate join venture, among other AI-related investments.
In exchanging an exposure to Nvidia, whose chips dominate the sector, for roughly 11 per cent of OpenAI, Softbank is also moving from an exposure to the “picks and shovels” supplier to the boom to the potentially more lucrative, but far riskier, more entrepreneurial research and development end of the sector, one with hefty capital expenditure requirements to train the large language models and develop the massive data centres that enable deployment of AI.
Softbank has more than $US300 billion of assets, but its commitments to OpenAI and Stargate and a $US1 trillion AI and robotics manufacturing hub in Arizona are stressing its financial capacity and liquidity. It has AI-related investment commitments of well over $US100 billion and Son has ambitions for an even bigger exposure to AI.
For Nvidia, whose shares fell per cent on the news, which some interpreted as a loss of confidence by SoftBank in its prospects, redeployment of the unlocked capital could actually be a positive if its improves OpenAI’s ability to finance its own $US1.4 trillion of commitments, which include more than $US100 billion of Nvidia’s chips.
That’s yet another example of how interdependent the major players in AI have become, and how circular the financial relationships between them are, with Nvidia and OpenAI at the centre of an increasingly dense web of relationships.
That incestuousness is being driven by the sheer scale of the demand for capital and the rate at which that demand keeps expanding. It is one of the biggest investment booms in history, with estimates that somewhere between $US5 trillion and $US7 trillion will be invested in AI-related infrastructure by the end of the decade.
The so-called “Big Five” hyperscalers – Amazon, Alphabet, Meta, Microsoft and Oracle – will alone spend more than $US375 billion this year on AI chips and infrastructure, more than $US500 billion next year and, on some estimates, more than $US600 billion in 2027, funded roughly half from their existing cash flows, but increasingly by external capital.
Startups like OpenAI, despite being valued at $US500 billion in recent equity raisings and planning a public float next year that could value it at $US1 trillion, don’t have any meaningful cash generation, relative to their spending, and are reliant on continually raising equity and, more recently, non-investment grade (and therefore quite expensive) debt.
Nvidia’s shares slumped on the news but the sale could turn out as a positive. Credit: Getty Images
The industry’s development is too expensive to be fully funded by the equity markets. The companies are going to have to tap into almost every corner of the global financial system to fund the acquisitions of chips, the building of the data centres and the power and water infrastructure required to operate them.
Banks are unlikely to be major direct and long-term financiers, given the risky nature of the underlying assets and their reliance on chips that have relatively short half lives, which requires continual reinvestment.
Private credit and private equity are already being tapped. Financial engineering, using off-balance-sheet vehicles, short-term data centre leases (to avoid having them classified as liabilities) with guarantees against loss and other “creative” techniques are emerging.
Vendor financing – ike the deal whether Nvidia agreed to invest $US100 billion in OpenAI in $US10 billion tranches that was matched by an OpenAI commitment to acquire $US100 billion of Nvidia’s chips, in similar tranches – is an increasingly common feature of the industry and one that highlights the interdependencies.
AI’s demand for computing power, or “compute” as the industry labels it, is astronomical, so the question of whether there is sufficient available financial capacity prepared to fund it is a significant one.
Similarly, there’s a question mark over whether development of the infrastructure – the data centres, power stations and water supply – can keep pace with the demand for compute.
JP Morgan recently estimated that about 122 gigawatts of global data centre installations would need to be built between next year and 2030, at an accelerating rate, requiring about 150 GW of power.
It takes three to four years to build a new gas-fired power plant and more than a decade for a new nuclear power station, so can the infrastructure build keep up with the industry’s needs and can those new sources of energy be funded when the credit exposure is effectively to a single industry that has yet to establish that its economics are sound?
The industry’s development is too expensive to be fully funded by the equity markets.Credit: Bloomberg
In the immediate future, the hyperscalers, with their strong cashflows and access to debt markets, can fund their share of the financing burden, albeit that it might entail cutting back on some of their non-AI-related investments.
Alongside equity for institutions, and some probably limited bank loans, investment grade bond markets, leveraged loan markets, securitisation markets and even government funding will need to be accessed.
The unknown is not whether there is demand for AI products, because there is, but whether companies and consumers will be prepared to pay enough to generate returns on AI investments commensurate with the risk.
There are now billions of AI users globally, but less than 5 per cent of the users of OpenAI’s ChatGPT – the dominant chatbot in the sector – pay for its services.
AI might be transformative, but its pioneers will crash and burn if they can’t generate the revenues and earnings to meet their commitments and do so within time periods and at levels that justify the scale and risk of their investments.
In the 1990s, telcos and cable companies invested massively in new networks whose capacity outstripped demand and users willingness to pay.
The telco and dot-com boom ended in a spectacular bust, but left a legacy of eventually useful infrastructure and intellectual capital as demand for that capacity grew. It also laid the foundations for today’s dominant tech companies.
That’s a possible outcome for the AI boom, albeit not an inevitable one for an industry that is still in its formative phase.
The capital constraints, relative to the demand for capital, and the questionmark over whether all the infrastructure required to meet the ambitions of all the industry participants can be built, will probably determine the industry’s near term outlook, winnowing the financially weak from the strong in the process.
The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.
Most Viewed in Technology
Disclaimer : This story is auto aggregated by a computer programme and has not been created or edited by DOWNTHENEWS. Publisher: www.smh.com.au
