Making a Market
"While full automation of the economy is presented here as an ideal and a demand, in practice it is unlikely to be fully achieved." - Srnicek, Inventing the Future
A primer, from Nick Land.
Last year I wrote a Substack outlining the thesis of data as a commodity, and computation as commodity processing, similar to an oil refinery refining crude oil as the commodity, I also outlined the meta thesis for companies that are working on the problem of commodifying data and computation, namely Valdi, a startup we are invested in at K-Capital and also a company I’m an advisor to (fair warning)
This blog serves as an exploration of products and services a company like Valdi can create and offer to computational markets that will:
drive the cost of computation towards zero;
Derive profitability by doing so;
Do so in a sustainable and self-reinforcing way.
Compute is already cheap.
Computation is relatively cheap.
And it has and will for the foreseeable future get cheaper over time.
Compute power per dollar has increased by a factor of ten every four years, over the last 30 years. So computation itself is not getting more expensive, however the amount we are computing is growing faster than the cost savings of hardware and manufacturing efforts.
For example, the Numerical Wind Tunnel, as a Japanese Aerospace Supercomputer used to compute fluid dynamics with a sustained compute performance of around 100 Gigaflops/second built in 1990.
It cost in 1990 an estimated $100,000,000.
Today a Nvidia RTX 4090 has a default compute speed of 83 Teraflops/s meaning a gaming graphics card anyone can buy from Best Buy costs $0.019 cents per Gigaflop.
We have effectively made compute 52,631,579 TIMES CHEAPER in under 35 years.
But we compute way more than supply allows.
Unfortunately, our computational load has grown tremendously in the same time frame.
In 2018, OpenAI found the amount of compute used in the largest AI training runs has been increasing exponentially with a 3.4-month doubling time (by comparison, Moore’s Law had a 2-year doubling period). Since 2012, this metric has grown by more than 300,000x (a 2-year doubling period would yield only a 7x increase)
This means we will be held back as the supply of compute is outgrown by the demands for computation into the known future.
So the problem remains a supply and demand problem, and generates a question: how can you accurately, quickly, and affordably price supply when its a forever compounding sellers market.
Jeff Bezos saw the internet growing at 2,300% a year when he was starting Amazon; computation has been growing at ~1000% a year since 2012, and accelerating.
Supply and Demand exist in a market.
The problem with things growing that fast is the difficulty to keep up.
Apple will be buying every 3nm chipset TSMC can produce;
Nvidia is sold out on A100s until Q2 2024;
This under-supply and over-demand will lead to a drive in cloud compute. People didnt want to own and manage their own servers and mainframes in the 1980s, they didnt want to build and service their own internal software in the 2000s, and they will not want to open a datacenter today.
We already see this with Amazon Web Services, Google Cloud, Microsoft Azure, and other cloud based computation solutions - but what are they missing?
Commodification.
All these cloud servers work as silos. You must pick who your provider is, and stick with them to integrate into their ever growing batch of services. But you’re stuck with them.
It’s like a marriage.
It creates a problem similar to mainframes, you are beholden to a fixed amount of compute power provided by your cloud provider, you pay for extensive overhead built into your fees, and you while you do have some flexibility in scaling up and down, at a certain point it becomes cost prohibitive on a per unit basis.
The solution is commodification.
Commodification in this sense means driving towards standardized, automated transactions that are priced at spot as the market moves.
Meaning buying compute right now, might be cheaper then tomorrow because of availability, or vise versa because you are buying from an open market, not a single provider.
Commodification means the ability to compare prices and undergo price discovery, it means ultimate flexibility in scaling.
Acceleration.
Commodification is great, and standardization is great too, but how do you actually make money from that?
It’s a given standardization gives way to cheaper pricing as you increase the supply of compute to users - but if things are cheaper how could someone capitalize and build a business around that?
Well the first step is building an open, interoperable market. Valdi is the best example of an open marketplace for compute that I have seen, but competitors like Render Network, Akash Network, and Vultr all claim to run open marketplaces.
However the next step is not so obvious.
The commodification of compute implies a market for computation, and a market for anything liquid also implies the ability to derive delivery contracts around said commodity.
The secret to Valdi and the long tail of what we have ideated, is a derivatives marketplace for computation.
Options, Forwards, Futures, Swaps, CFDs; all these things in a financial market for a commodity act as price stabilization and liquidity providers. They also will exist for computation.
Without giving too much away the long tail of Valdi is not just a computational resell shop operating under the behemoths of Amazon and Microsoft, its the financial commodification and engineering of a new market for computation.
One where pricing and supply is not opaque, one where computation is more fungible than money, and one where Valdi can generate profits off of technical and financial services provided on their marketplace, similar to a CME or LME; along with the ability to trade in the commodity of the future, compute.
The good news for us, is that much of the financial engineering has been done with the traditional financial system - we’re not trying to reinvent the wheel here - we’re looking to take concepts that have generated trillions of dollars in wealth across all of human economics, and apply it to a new type of commodity - computation.