In the context of computing power deflation, Nvidia is continually launching new machines, from H200 to GB200. Once sold, it becomes fuel for $NVDA's market capitalization, transferring the pressure to monetize to the machine holders. In this situation, the most valuable things that crypto computing power project parties can do will be:
Introduce more high-end computing power to play the role of mining head.
Sell orders that have already generated cash flow at a discount to recover principal in advance, serving to reduce long-term leverage.
Utilize its own network effect to link idle computing power with clients.
Let's analyze Aethir's recent efforts based on these three points:
Introduce high-end computing power
Aethir's recent two developments are New Horizon and deep cooperation with GAIB & GMI Cloud.
The goal of New Horizon is to attract more high-end GPU holders to join Aethir, holding more callable machines in various countries and regions. From the application process on the official website, the machines being introduced are mainly gaming graphics cards above RTX 3080, as well as data center-grade graphics cards like H100, H200, and even GB200 NVL2. According to the documents, Aethir Labs has provided the corresponding computing power reward coefficients, likely based on current market memory and floating-point performance tests. Collaborating with GAIB & GMI Cloud brings in H200 machines; GMI Cloud is Nvidia's official cloud service partner and has received investment from Supermicro, being one of the first service providers in the Asia-Pacific region to obtain H200. Therefore, Aethir also becomes one of the first project parties to introduce H200 into decentralized computing networks.
According to the current computing power market, the H200, with its memory capacity of up to 141G and memory bandwidth of 4.8T/s, is nearly twice that of H100, which helps to train larger parameter models more quickly. Therefore, AI large model companies are also willing to pay a premium. From this perspective, it is more conducive to hedge the risk of other low-end computing power price declines.
Recover principal in advance
Aethir's decentralized cloud includes over 3,000 NVIDIA H100s and more than 43,000 consumer-grade graphics cards, some of which are owned machines. Facing the impact of new models and the shrinking demand from AI companies, it is necessary to issue corresponding revenue products to recover the principal of machine purchases in advance. GAIB can tokenize GPUs through the financialization of computing power, allowing them to circulate further in the DeFi system. This is essentially a win-win operation, bringing external underlying revenue assets to the DeFi system while allowing GPU machine holders to recover their principal early to purchase new machines and expand scale advantages.
Network effect
From the perspective of the economic model, Aethir's high-end machines require B2B access, needing to fill out forms and sign contracts with Aethir Lab, which then prices based on region and machine type. Here, Aethir can fully leverage the network effect, using its platform to seek AI clients, while also charging service fees and price differences to GPU providers, generating positive cash flow.
The positive aspect here is that machine providers first need to purchase $ATH for staking; the service revenue generated by the machines needs to pay a 20% service fee to the Aethir network, issued in the form of ATH. The more high-end machines there are, the larger the buying demand formed. Taking H200 as an example, according to the formula provided by Aethir Lab, an eight-card H200 requires staking 1,469,235.0552 $ATH, which, based on the market price, is close to 73,000 USD, essentially reaching 25% of the machine's value.
Summary
In short, the subsequent computing power industry has a high Matthew effect, which corresponds similarly to crypto computing power projects. If there is no GPU physical business at hand, that might be fine, as there will be no losses; the FDV will just revert from a machine-supported valuation back to Meme levels. If there is a physical business, the only way is to move upwards, using high-end machines to lock in selling pressure while generating revenue to combat the computing power deflation caused by NVIDIA's market capitalization increase.