From The Economist, February 17:
Get ready for a new wave of securities, hedges and collateral
This year just five American tech giants are set to make $700bn-worth of capital expenditure, as investment in the data centres needed for artificial intelligence surges. Investors have long quipped that “data is the new oil”; now they are backing firms to spend far more to process it. By comparison, the oil and gas industry invested just $570bn in exploration and production last year.Yet in another sense, data—or at least the chips on which it is stored and manipulated—still lags far behind oil. Graphics processing units (GPUs) play a tiny role in financial markets. True, some loans use them as collateral, but they remain hard to price and to sell on. There is practically no market for GPU derivatives, which allow traders to parcel up and offload risks (such as that of a price crash). Chips and the processing power they generate (or “compute”) are therefore ripe to be financialised by Wall Street—just like oil, housing and myriad other assets before them. A wave of new innovators hopes to do just that.
Take OneChronos, a fintech firm established in 2016 to make share-trading more efficient. The company aims to launch a market for compute, on which bundles of goods can be auctioned, by June. It has paired up with Auctionomics, co-founded by Paul Milgrom, who won the Nobel prize for economics in 2020. Then there is Ornn, which has launched an index tracking the prices of chips, including Nvidia’s popular H100. The startup also plans to sell put options—derivative contracts that pay out if prices fall sharply—on physical GPUs.
Eventually, the combination of trusted benchmarks and liquid derivatives markets could support bonds collateralised by baskets of GPUs, much like those backed by bundles of individual loans. Few investors possess detailed knowledge of Californian consumers or Ohio’s office market. Many nevertheless snap up bonds collateralised by credit-card debt and commercial mortgages.
To build such a market, some fearsome obstacles must be overcome. Perhaps the biggest one is that the most advanced chips tend to lose their value fast, as even whizzier ones come to market. Morgan Stanley estimates that this could prompt Alphabet, Microsoft, Meta and Oracle to record depreciations worth $680bn over the next four years. Any estimate like this is subject to huge uncertainty, since no one can know for sure how the technology will progress. This represents a big risk to buyers of loans collateralised by GPUs that might unexpectedly tumble in price. Even now, for the biggest buyers of chips, comparing the most recent models to those of a decade ago is like comparing a supersonic jet to a horse and cart.
Then there are the difficulties with trading compute. Users need to be fairly close to their data centres, which, unlike fuel, cannot be shuttled from one place to another. And so prices vary a lot from region to region, since it is not easy for trade flows to match supply to demand.
If such obstacles could be overcome, however, the prize of more sophisticated financial instruments would be a hugely valuable one. Derivative contracts could help allocate risks—that particular types of chip suddenly become worthless, for instance—to traders who actually want to take them, in exchange for a premium. That might allow companies with business models which depend on GPUs to worry less that their assets might rapidly become obsolete and focus more on improving their operations. Conversely, compute-hungry startups would find it easier to borrow if their loans could be collateralised with chips, then bundled together with others and sold as bonds....
....MUCH MORE
