In 2024, the field of computing power investment shifted from being extremely popular to a state of calm. The supply of paper computing power in computing power investment is severely plagued by bubbles. As the end of the year approaches, negative news emerges. Here are two phone conversations that reflect the changes in computing power investment.
The first phone call is from a high-level executive of an energy enterprise. In the past year, they were actively consulting about computing power investment. Initially, they were misled by an intermediary into investing in computing power to obtain new energy quotas, but later gave up due to risks. The reasons include: IT assets depreciate quickly, the asset value of equipment in commercial operation is uncontrollable, and it is easily defined as loss of state-owned assets; new energy quotas are supplemented later and there are risks during the construction period. The top leadership did not give a clear directive for investment. The intermediary’s statements were half-understood, and the project’s benefits cannot enter one’s personal pocket. There’s no need to risk one’s personal future.
The other phone call is from an old colleague who is responsible for infrastructure in an internet company and is facing difficulties in renting computing power. The company plans to rent 10 units of H Yāolínglíng at 68,000 yuan per month for a one-year contract but cannot find the resources. The reasons are as follows: First, someone made a promise but didn’t fulfill it. Due to the investigation by the US Department of Commerce on Nvidia, they can’t get the goods. Second, the demand is called a “small order” (short lease period, low rent, and small usage).正规经营的算力租赁方不愿接正规ly operated computing power lessors are reluctant to take it. The specific explanations are as follows:
1. Rent issue: It’s impossible for customers to sign a short-term contract at the price of a large customer’s long-term contract. The service provider needs to sign a long-term contract based on the current investment to ensure revenue. There are two types of one-year contracts. One is suitable for small demands but has a high price. The other is for large customers who sign a batch every year, and each batch is signed for more than three years.
2. Usage issue: 10 units may be a lot for the customer, but it may be very little for the service provider. Renting machines is different from renting computing power. The cost of renting computing power includes machines, networking, IDC cabinets, electricity, labor costs, and reasonable profits. Taking a cluster of 128 units as an example, each switch is connected to 32 servers. Renting only 10 units will waste two-thirds of the switch cost. The service provider hopes to provide in multiples of 32 units. Although some people have successfully rented only four units, it is mostly in specific circumstances and requires luck.
3. Cost calculation: The one-time investment of the service provider divided by the expected number of months for fund recovery is the monthly cost. Taking a single server with a purchase cost of 2.3 million yuan as an example, plus the supporting IB network accounting for 13%, spare equipment cost accounting for 3%, and capital cost accounting for 10%, the one-time capital cost investment for a single machine is 2.9446 million yuan. Spread over 60 months, it is 49,000 yuan per month. But there is also an IDC cost of 7,500 yuan per unit, business expense of 1,000 yuan per unit, and operation cost of 800 yuan per unit. After 60 months of “hard conversion to service” for a server costing 2.3 million yuan, the average monthly cost reaches 58,300 yuan. If there are vacancies in the cluster or the lease period is short, the cost is even higher. Signing a 3+2 contract is beneficial for the service provider to quickly recover funds in the first three years. If only signing for three years, the cost bottom line of the service provider needs to be considered.
The current situation has changed: Intelligent computing centers and computing power leasing businesses are essentially cloud computing businesses. Personnel from non-ICT industries such as traditional enterprises, trade, finance, and new energy have entered, inflating the supply bubble. Exaggerated delivery capabilities and excessive promises have led to many contract violations. The government’s investment situation is uncertain. In some places, non-commercial operations may lead to consumption problems. State-owned enterprises and central enterprises also face many problems when participating in computing power investment. Large customers may mislead people who don’t understand accounting. Currently, there is a situation of “paper supply surplus but insufficient effective supply” across the country. It is recommended to seize the opportunity to negotiate and sign while computing power can still be rented. Otherwise, the price
