Interruptible Data Centers Increase Reserve Margin, Optimizing Power Assets
BitOoda is Leading The Way on the Financialization of Compute and Power Asset Portfolio Optimization
The global demand for compute power is growing at an exponential pace, driven by the demand for AI applications, machine learning, computational biology and other emerging technologies. ChatGPT has exposed the massive computational load of AI to the general public, but the underlying trend has been clear for a number of years.
At the same time, blockchain-related compute is seeing increasing complexity. The Proof of Work (PoW) paradigm in crypto needs large amounts of efficient compute, as the first to solve a puzzle earns the revenue. While Proof of Stake (PoS) products reduced the immediate compute need, scaling and privacy applications such as Zero Knowledge Proofs (ZKPs) require large amounts of cutting edge computing power. This is an important distinction. GPU-based PoW could be solved with mid-range hardware deployments that could not meet the demanding requirements for AI algorithms. However, ZKPs are extremely resource-intensive, and require some of the same high-end hardware that advanced compute applications need.
This is driving a convergence of blockchain compute and general purpose compute with tremendous implications for the entire data center industry.
Today, most data center operators secure contracts with either large end users such as Pfizer or Ford or Citibank, or large cloud providers such as Google Cloud, Microsoft Azure or Amazon AWS. They accept a relatively low gross margin for the stable revenue, high-capacity utilization and the customer acquisition cost savings that come with selling to cloud providers, who in turn make significantly larger margins selling to end users.
The convergence of blockchain compute with advanced general purpose compute supports independent data center operators, because blockchain applications such as ZKPs can provide the data center with a baseline revenue and can drive capacity utilization that can fund the business as it seeks end customers. The data center now becomes a “compute refinery,” dynamically orienting itself to changing customer applications, producing the compute that provides the highest value to it, and switching to lower value base load for idle capacity. In this model, there is always some revenue.
There is another convergence happening simultaneously: data centers consume massive amounts of power – north of 50GW a year, and growing rapidly. A data center can switch off operations rapidly when the grid needs a flexible load and can be compensated well for this controllable load – provided the load is interruptible – without consequences that a mission critical application cannot tolerate.
As the US and the world moves to more renewable energy, the old paradigm of supply needing to respond to a variable load is shifting toward a paradigm of the load needing to respond to a variable supply. This encompasses seasonal, hourly and unpredictable or short lead-time variances.
This is a key challenge for our grid infrastructure, particularly over the next decade or so while battery deployments remain modest. In the future, when utility and grid scale battery backup becomes ubiquitous, along with upgrades to long distance transmission line capacity, the electric infrastructure system will once more be able to modulate supply to meet variable demand, but demand response will likely remain a long term part of the solution.
Unlike most heavy manufacturing or process industries that cannot modulate demand, an interruptible data center can. This can increase the theoretical reserve margin, since at peak load there will be load shedding from datacenters, giving a buffer to the system. Such an insurgent data center could compete with incumbent data centers on price, while delivering true green credentials through helping stabilize the grid and optimize the power asset portfolio.
This will form the basis of the great convergence of Compute, Blockchain and Power to offer true fungible compute for the first time. The parameters of such compute capacity would be standardized in terms of compute amount (say, petaflop hours or PFHr) and the interruptibility vs mission criticality. Other parameters such as the hardware configuration would also see standardization, because if the buyer is paying $x per PFHr and can receive the verifiable result by their deadline, they are indifferent to whether the service provider utilizes one or another specialized configuration.
The key missing piece remains a marketplace for fungible compute where sellers can offer their capacity and buyers can bid on it. This marketplace will eventually correlate with power and even carbon markets, as compute producers and consumers seek to hedge against the biggest operating expense for a data center – power – and power providers optimize the delivery of compute across a data center portfolio that allows them to shift demand across both time and space, optimizing their power generation assets and delivering grid-scale stability and efficiency benefits. Eventually, we see the incumbent cloud providers joining the insurgent data centers in this endeavor – primarily owing to end-customer demand pull.
We at BitOoda are partnering with a variety of constituents to pioneer a suite of products and services to develop the compute marketplace – including transparent pricing, a matching engine between buyer and seller, and smart routing of data across sites while maintaining a robust chain of custody for data integrity and privacy, and the underlying set of hedging tools to manage risk while eliminating opaque middlemen. This critical emerging market demands a structured, transparent infrastructure based on traditional capital markets principles.