GPU Finance

Problems with Decentralized Cloud Services

De-Pin has presented the potential to solve various problems of centralized networks. It is cost-efficient, environmentally friendly by using idle resources, and excellent in terms of security and stability. Nevertheless, several issues have arisen as De-Pin services have gone live.

The biggest issue is the unstable profitability for device providers. Providers have to wait for their devices to be rented (Hired) by someone. Although the De-Pin service protocol automatically assigns rental tasks, irregular rentals make it difficult for providers to predict their profitability. This process can lead to an imbalance in rental assignments, and it is hard to understand why unfair rental assignments continue. An opaque protocol creates misunderstandings and dissatisfaction among participants, hindering their continued participation.

Looking at the general operation of De-Pin, device providers are rewarded with tokens just for connecting their devices to solve these problems and ensure a stable supply. However, if tokens are given to meet the expected profits of providers, it can lead to token inflation, making the expected profits poor. Initially, during the early stages of the service, device purchase and operational costs can be quickly recouped, but continued price decline due to token inflation makes it difficult to attract new providers and retain existing ones.

This problem is a significant challenge not only for device providers but also for the project team (foundation). If the tokens, which are the primary fuel for operating the protocol, are distributed indiscriminately and lose their value, the protocol's credibility will also be lost. Token inflation issues are common in Web3 projects, leading to frequent observations of services losing momentum and plummeting.

How Does Amazon Cloud Make Money?

Before explaining the solution, it is necessary to understand how traditional clouds make money. As of 2024, the hourly usage fee for NVIDIA A100 GPUs in existing cloud services is about $4. If 1,000 NVIDIA A100 GPUs are rented for a year, the revenue generated is $35,040,000. It costs about $20,000 to purchase one NVIDIA A100, and $20,000,000 to purchase 1,000 units. The return on investment compared to the purchase cost of the product is 175.2%.

If the annual power consumption of 1,000 A100 GPUs is calculated at the average industrial electricity rate in the United States ($0.08/kWh), it amounts to $277,000. Considering other operating costs, security control costs, power consumption, and electricity rate increases, the cost of sales is minimal compared to the revenue of $35,040,000, leaving a substantial net profit.

Expected Returns for GPGPU Participants

Assuming GPGPU charges $2 per hour, a 50% discount from the $4 hourly rate for NVIDIA A100s, the annual revenue for renting 1,000 units would be $17,520,000, with an annual return of 87.6% compared to the cost of purchasing new equipment. To cover the $277,000 electricity cost, the annual expected return needs to be only 1.66% to avoid losses.

  • Discount compared to traditional cloud: 50% ($4 → $2)

  • Annual return rate with 70% GPU availability: 61.32%

This calculation shows that even with a substantial 50% discount and only 70% of the total equipment in operation, an annual return rate of 61.32% can be expected.

Assuming 1,000 NVIDIA A100 GPUs are supplied through a decentralized GPU cloud and an annual return rate of 30% is proposed, GPU providers can achieve a high annual return rate even after deducting electricity costs. If someone distributes stable returns to GPU providers with their own money in exchange for higher future profits, both GPU providers and the "someone" can achieve an ideal structure where both parties are satisfied. GPGPU has solved this problem with the concepts of a flexible pool and GP Bond.

GP Bond is an investment means for companies and individuals to invest in a flexible GPU Pool voluntarily and earn profits. GP Bond allows for easy entry and withdrawal of investments anytime, making it easily liquid when funds are needed. This is very useful in emergencies or sudden investment opportunities, allowing for stress-free investment and thus attracting significant funds.

Additionally, GPU providers can receive rewards for their connected time without waiting for their GPU to be rented. They can achieve stable and predictable returns regardless of rental status.

Flow of Investment Funds

1. Investment Injection

Investors deposit funds and receive GP Bonds. GP Bonds are tokens with intrinsic value based on GPU resources and are proportional to the market value (annual rental cost) of the entire GPU Pool. GP Bond holders can share in the profits of the GPU Pool. GP Bonds have a nominal value equivalent to the invested funds and, unlike general governance tokens, do not have price volatility.

2. Use of Investment Funds

Investment funds are segregated and stored in a separate contract and used as rental fees for the flexible GPU Pool. Rental fees are calculated in minimum units of one hour and settled every 24 hours. Providers connected to the GPU Pool (providing GPUs) are rewarded based on the annual rental yield after maintaining a minimum uptime of 7 days.

3. Yield Calculation

The yield is determined by three factors. The first is the actual revenue generated by renting the GPU Pool, the second is the annual return for Bond investors, and the third is the annual rental fee of the GPU Pool. The remaining revenue after subtracting the GPU Pool rental fee from the revenue generated by renting the GPU Pool to users is distributed to the GP Bond held by Bond investors.

  • GPU rental revenue (α)

  • Bond investor yield (β)

  • GPU Pool rental fee (γ)

  • Margin gap (δ)

  • Service fee (ε)

Margin gap (δ) = GPU rental revenue (α) - GPU Pool rental fee (γ)

Bond investor yield (β) = margin gap (δ) - service fee (ε)

Therefore, there must be a certain gap for paying rental fees to the GPU Pool between the GPU Pool rental revenue and the yield for Bond investors, referred to as the ‘margin gap’. The margin gap is determined by the supply and demand of GPUs.

Payment to GPU Providers

  • GPU Pool equipment scale: $100 million

  • GPU Pool annual rental fee: $30 million

  • GPU Pool annual rental yield: 30%

Revenue from GPU Rentals

  • Assumed GPU Pool availability: 70%

  • Assumed GPU Pool yield: 61.32% ($61.32 million, assuming 1,000 A100 GPUs)

  • Margin gap: 31.32% ($30 million)

  • Service fee: 3%

  • Bond investor yield: 28.32%

Adjustable variables here are the rental fees paid to the GPU Pool, the margin gap, and the service fee. By adjusting the annual rental fee proposed to the GPU Pool, more GPUs can be supplied, or the scale can be reduced. The margin gap can be increased or decreased to adjust the scale of Bond investments. By adjusting the variables according to market conditions, the balance between GPU Pool supply and Bond investments can be maintained.

3. Redemption of GP Bonds

Bondholders can redeem their bonds to recover their investment funds. GP Bonds have a maturity of 365 days from the time of issuance, and the rental fee for the GPU Pool is deducted daily from the invested principal. Additionally, the revenue from the GPU Pool is distributed daily. Therefore, the recoverable investment funds are calculated as follows:

Recoverable investment funds = investment principal + GPU Pool rental revenue - GPU Pool rental fee

As long as the GPU Pool's availability yield does not fall below the rental fee, the possibility of investment loss below the principal is low. An automated protocol operates to maintain the availability yield above a certain level to ensure sufficient margin gap. If availability is low and sufficient margin gap is not maintained, lower quality GPU devices will be automatically removed from the GPU Pool, ensuring higher availability and reducing GPU Pool rental fees.

4. Secondary Trading of GP Bonds

GP Bonds can be traded between GPGPU service users. The face value of the token does not change, and a premium can be added for the remaining period until maturity. The premium is autonomously set and traded.

The price of GP Bond = face value + premium

Trading is possible on the GPGPU website and must be conducted by users who have completed KYC. Customer verification procedures are essential to comply with anti-fraud and anti-money laundering obligations. Unauthorized trading on DEXs or centralized exchanges is not allowed.

GP Bond Protocol Scenarios

Increase in GPU Availability

  1. Increase in GPU availability: GPU usage demand increases or GPU providers leave, increasing overall GPU Pool availability.

  2. Increase in Bond yield: If usage demand increases, Bond investment yield also increases, and if GPU providers leave, the GPU Pool scale decreases, lowering rental fees, thereby increasing Bond investment yield.

  3. Inflow of Bond investment funds: As Bond yield increases, STO investment funds increase.

  4. Stabilization: As new Bond investment funds flow in, Bond investment yield and GPU availability stabilize.

Decrease in GPU Availability

  1. Decrease in GPU availability: GPU usage demand decreases or GPU providers join, decreasing overall GPU Pool availability.

  2. Decrease in Bond yield: If usage demand decreases, yield also decreases, and if GPU providers join, the GPU Pool scale increases, raising rental fees, thus decreasing Bond investment yield.

  3. Withdrawal of Bond investment funds: GP Bond holders redeem and exit.

  4. Reduction of GPU Pool: GPGPU reduces the scale of the GPU Pool, or GPU providers leave due to the withdrawal of Bond investment funds.

  5. Decrease in GPU rental fees: The scale of the GPU Pool decreases, reducing overall rental fees.

  6. Stabilization: Bond yield increases again, leading to stabilization.

The inflow or withdrawal of Bond investment funds, the increase or decrease of the GPU Pool is determined by GPU usage demand (availability) and is excluded from the scenarios.

Last updated