As the demand for AI inference transitions from the research confines of laboratories to the expansive realm of industrial applications, the landscape of computing power supply is quietly undergoing a revolutionary shift. Personal computers are being transformed into vital computing nodes, the idle computing resources of internet cafés during off-peak hours are being harnessed, and the latent computing power on cloud platforms is being meticulously scheduled. Amidst this transformative wave, how has a startup, established merely two years ago, leveraged a platform-based model to reshape the computing power market? On November 4, Wang Peng, the Co-founder of Gongji Computing Power, graced the Jixin Enterprise Services live broadcast room. He shared insights into how the 'Computing Power Didi,' a brainchild of Tsinghua University's laboratory, is bridging the gap between supply and demand, making computing power as ubiquitous and accessible as water and electricity.

This dialogue delves into the fundamental tenets of platform-based entrepreneurship:

Model Innovation and Barrier Establishment: What drives the choice of the shared computing power sector? How does technical scheduling overcome stability hurdles?

Supply-Demand Equilibrium and Growth Catalysts: How to navigate the 'chicken-and-egg' predicament? When does the tipping point occur?

Organizational Evolution and Risk Mitigation: How does a 50-member team sustain its entrepreneurial vigor? What strategies are in place to avert cash flow crises?

Will there be an explosion in AI inference demand? Can the asset-light model withstand the risks of hardware evolution? The answers are embedded in this dialogue.

Left: Wang Peng, Co-founder of Gongji Computing Power; Right: Jiang Wen, Founder of Jixin

Wang Peng offered the following key insights:

'Leveraging shared idle computing power and adopting elastic per-second billing can slash computing power costs by one-third.'

'Platform-based entrepreneurship is not a zero-sum game but an art of maintaining dynamic equilibrium.'

'Startups should position themselves as the 'agile monkey in the middle,' identifying a niche within their capabilities.'

Below is a curated selection of highlights (watch the full live broadcast replay on 'WeChat Channel: Jixin Enterprise Services'):

Jixin: Could you provide a brief introduction to Gongji Computing Power, Mr. Wang?

Wang Peng: Thank you, Mr. Jiang. It's a pleasure to engage with everyone. I'm Wang Peng, Co-founder of Gongji Computing Power. Our entrepreneurial journey commenced in 2023, and I'm currently pursuing a Master's in Engineering Management at Tsinghua University. My foray into the AI industry began in 2015, and my entrepreneurial aspirations, which had been simmering, finally materialized in 2023. The core brand of Gongji Computing Power is 'Suanleme' ('Is It Computed?'), focusing on shared idle computing power. We've constructed a platform akin to 'Computing Power Didi,' connecting the dots between supply and demand.

On the supply side, we encompass personal computers (which, after installing our computing power network client and agreeing to accept orders, can undertake tasks like AI image generation and scientific computing), internet café computers (with significant idle time during off-peak hours), and idle spot resources from data centers and cloud providers.

On the demand side, we offer flexible, elastic, and pay-as-you-go services, with per-second billing options. Many of our clients are AI enterprises, catering to their dynamic computing needs.

Technologically, we rely on cutting-edge computing power scheduling technology to facilitate the seamless entry and exit of computing resources. This technology, currently at the forefront domestically, is backed by the technical prowess of our core team from Tsinghua.

01 A Lone Demand Sparks the Sharing Economy

Jixin: What inspired Gongji Computing Power to envision a shared platform for the computing power industry? How did this idea germinate before the release of GPT-3.5 in March 2023?

Wang Peng: The entrepreneurial spark was ignited by the real needs of our founder, Fu Zhi. In 2022, Fu Zhi faced a deadline crunch while trying to publish a paper, with his computer estimated to take 40 days to complete the computational tasks. With only a week remaining, he explored multiple cloud platforms but found flexible leasing services either unavailable or prohibitively expensive. In a desperate move, he borrowed eight computers from his seniors and completed the calculations in under five days. Realizing that other students might encounter similar challenges, he built a rudimentary web-based shared computing power platform.

The platform unexpectedly gained traction on Bilibili, crashing the server that night and generating over 200 genuine orders within a month. This prompted Fu Zhi to consider commercialization, leading to extensive market research to ascertain if demand was widespread, if the supply side was adequate, and if technology could support efficient scheduling. The research affirmed the project's viability, with the core breakthrough being technical scheduling capabilities. We drew inspiration from virtual power plant scheduling strategies to address stability issues.

Following ChatGPT's release, we identified the inference side as more promising, given its short-lived tasks, high demand, and role in the commercialization monetization chain. Currently, this direction seems prescient: inference demand is widespread and substantial, with ample supply-side resources. Some clients even serve as both demanders and suppliers.

Jixin: Entrepreneurship is often likened to traversing a U-shaped tunnel, with the middle section being the most susceptible to confusion. How should entrepreneurs navigate the decisions between 'innovating or not' and 'practicing or not'?

Wang Peng: Innovation spans a spectrum from minor tweaks to revolutionary changes. Major innovations necessitate careful consideration of costs, including the capital chain, and should not be pursued recklessly.

To embark on the entrepreneurial journey, one must practice while keeping trial-and-error costs in check. After graduating, I spent three years at a state-owned enterprise before gaining experience at a startup. Before venturing out on your own, extensively study entrepreneurial cases. If you haven't found a compelling direction, consider joining a startup or collaborating with someone who has a promising idea.

02 Building Pathways Amidst 'Chickens and Eggs'

Jixin: With fierce competition on the AI application front, how can startups avoid being overshadowed by large corporations? Where do the barriers and moats lie?

Wang Peng: Barriers can be constructed in several ways:

Possessing scarce resources, such as vertical domain data or specialized client relationships;

Founders in the To C space must accurately gauge user needs and rapidly iterate products. Startups have an edge over large corporations in terms of iteration speed.

We advocate the 'Monkey Theory': Startups should position themselves as the 'agile monkey in the middle,' selecting areas within their capabilities and less fiercely contested. For instance, our partner in the internet café industry outmaneuvered giants by rapidly iterating products and maintaining high-density client communication, achieving favorable growth.

Jixin: What is the paramount challenge for platform-based entrepreneurship? How can it be overcome?

Wang Peng: The biggest hurdle is the supply-demand balance issue, often referred to as the 'chicken-and-egg' dilemma, while ensuring growth and maintaining stable income and profitability. The current financing environment lacks massive capital infusions, increasing the difficulty.

The supply-demand balance issue essentially requires operating like a 'dynamic balance master' to avoid scenarios where demand is abundant but supply is insufficient, or supply is connected but demand is lacking.

Additionally, cash flow must be robust to ensure rapid scaling during demand surges.

Jixin: Platform-based entrepreneurship faces an inflection point when the bilateral network hasn't formed, and the platform isn't truly established. How does Gongji Computing Power assess and manage this inflection point?

Wang Peng: Different platforms exhibit varying dynamics. Our business model achieves win-win outcomes: idle computing power from personal computers generates value through the platform, while demanders obtain elastic computing power at competitive prices. Our core competitiveness lies in technological leadership and team efficiency. Drawing from Meituan's playbook, its strong offline expansion capabilities and high organizational efficiency were key to its success. We'll continuously refine our products, enhance experiences for both sides, and leverage AI tools to boost organizational collaboration efficiency.

03 Sincerity: The Ultimate Weapon

Jixin: How has Gongji Computing Power enhanced organizational efficiency as it expanded from a small team to 50 members?

Wang Peng: We focus on three main pillars:

Personnel recruitment: We rigorously screen candidates, prioritizing 'self-igniting' talents with an entrepreneurial spirit;

Cultural construction: We foster a transparent, mutually supportive atmosphere, allowing employees to feel the team's positivity;

Goal management: We implement the OKR model, involving employees in goal setting to stimulate initiative.

Jixin: What are Gongji Computing Power's mission, vision, and values? How were they formulated?

Wang Peng: The essence of our mission and vision is 'Uniting Thousands of Households, Co-creating Wisdom, Benefiting the Future,' stemming from our initial consensus on 'inclusive computing power.' For instance, a Tsinghua classroom had over 200 students needing 4090 graphics cards for classes but lacked sufficient resources. We resolved this by integrating idle computing power from the C-end.

Our values emphasize integrity and candor, but the key lies in practice rather than mere slogans. Partners must lead by example, communicating openly in both work and personal interactions.

Jixin: In the event of a cash crisis where salaries can't be paid, what are the core team's deliberative principles?

Wang Peng: Founders must exercise prudence in cash flow management, setting a cash flow countdown to preemptively avoid funding issues;

Additionally, based on the principle of 'candor,' we first transparently communicate the situation, solutions, and efforts with core members. Simultaneously, we understand that some employees may leave due to personal circumstances and avoid being overly critical.

04 Navigating Technological Evolution with an Asset-Light Approach

Jixin: Can Gongji Computing Power's scheduling platform address local computing power challenges in underdeveloped countries and regions? What is the progress of overseas expansion?

Wang Peng: Such platforms hold immense potential in underdeveloped areas for educational and public welfare purposes. Regarding overseas expansion, we've established a branch in Hong Kong. This year, our primary focus is on the domestic market, with plans to accelerate overseas business expansion next year.

Jixin: NVIDIA graphics cards undergo rapid iteration, and computing power investments must account for discount rates and residual value depletion. How does rapid infrastructure iteration impact project return rates?

Wang Peng: Rapid iteration of new cards affects investment returns and necessitates a rational response. Computing power investments demand professional acumen to ensure competitiveness over a 3-5 year horizon. We adopt an asset-light model to mitigate hardware investment risks while insisting on 'staying in the game' to maintain industry relevance. For inexperienced investors, we recommend collaborating with professional institutions.

Jixin: Now, let's enter the 'Message in a Bottle' segment. Please respond to the question posed by the previous guest, Liu Changwei, Founder of Yixin Technology: Are you currently profitable and earning money from AI?

Wang Peng: I'll address this from two perspectives:

Personal level: In 2023, I established a paid community, uploading AI-related materials. Users recharged, and although the amount was modest, I generated revenue by sharing knowledge.

Company level: The better AI evolves, the greater clients' computing power demand, and the higher our revenue. Some quarters have been profitable, but overall profitability this year is challenging due to substantial platform construction investments.

Jixin: Please pose a question for the next guest visiting the Jixin Meeting Room.

Wang Peng: My question is: Have you noticed a change in your weight since embarking on your entrepreneurial journey? If so, how did you address it?