How patient capital can accompany the AI industry through R&D cycles

At the opening ceremony of the 2026 World Artificial Intelligence Conference and the High-Level Meeting on Global AI Governance, President Xi Jinping pointed out: “Artificial intelligence is a new engine of world economic growth and an accelerator for the transformation of old and new growth drivers, and it is moving from the ‘digital world’ to the ‘physical world.’” For AI to go from algorithmic breakthroughs to large-scale commercial use, it must go through long-term R&D and pilot validation. Large-model training and computing power construction require enormous investment while returns lag, and their asset-light, high-risk characteristics make it difficult to rely on traditional credit. This technological long march is also a capital long march. Patient capital focuses on long-term returns and tolerates trial and error; it can accompany enterprises through R&D cycles and industry downturns, and is naturally suited to the growth patterns of the AI industry. To strengthen patient capital and guide long-term money into long-term investment, we must conform to this suitability, broaden sources of long-term funds, improve fault-tolerant incentive arrangements, smooth capital circulation channels, and guide capital to invest early, in small firms, for the long term, and in hard technology, thereby injecting lasting momentum into the high-quality development of the AI industry. Recognize the intrinsic fit. Based on the laws of growth, recognize the intrinsic fit between patient capital and the AI industry. Artificial intelligence is a typical hard-tech track. For a technology to go from a laboratory algorithm to large-scale commercial use, it must pass through multiple stages such as basic research, prototype development, pilot validation, and market cultivation. Large-model training and computing cluster construction involve investments measured in hundreds of millions of yuan, yet the technical route may be rewritten by a new paradigm within a year or two, and enterprises must “burn money” for a long time before becoming profitable. Global technological competition also waits for no one. Once financing is interrupted, the technology window closes in an instant. Many enterprises fail not because of their technical route, but because their cash flow cannot be sustained. At the same time, the most valuable assets of AI enterprises are algorithms, data, and R&D teams; they lack collateral such as land and factories. Traditional credit values collateral and stable cash flow, and when faced with financing needs in the AI industry, it often dares not lend or is unwilling to lend. Social capital seeking short-term arbitrage is accustomed to chasing rises and dumping falls, cashing out and exiting at valuation peaks, so capital supply and the rhythm of the industry are clearly misaligned. Patient capital can precisely fill this gap. It matches long-cycle R&D with long-term funds, does not require annual dividends or quarterly repayments, and can accompany enterprises through the start-up period with no revenue. It shares risks through equity investment, invests early, in small firms, and in hard technology, and has a higher tolerance for failures in technological exploration. It can also bring industrial resources beyond funding, helping start-ups connect with computing power, data, and application scenarios, and introducing standardized governance and follow-on financing. By contrast, short-term capital amid hot-spot rotation rushes in when the market is hot and withdraws when the industry cools, thereby amplifying industrial fluctuations. Terms match, risks can be borne, and growth can be accompanied—this is the root of the fit between patient capital and the AI industry. Broaden the source of living water. With broadening the source of living water as the lever, strengthen long-term capital invested in AI. At present, the total amount of funding supporting AI is not small, but its structure is not ideal. Venture capital funds have relatively short durations and often must liquidate and exit before hard-tech enterprises mature. Genuine long-term money such as social security, insurance, and annuities still has limited allocation to venture capital. Funds crowd into later-stage projects with clear business models, while seed-stage and start-up AI enterprises find it hard to raise funds. Some local guidance funds also suffer from dispersed scale, short-term-oriented assessment, and reluctance to invest in early stages. To resolve the contradiction between short-term money and long-term investment, we must start from the source of funds. First, expand the ranks of long-term investors and implement policies for investment by social security funds, insurance funds, and enterprise annuities, steadily increasing allocation ratios under the premise of prudent regulation. Optimize fund duration design, encourage the establishment of long-term funds of more than ten years, and implement cross-cycle assessment for long-term investment so that investors dare to commit money for the long haul. Second, government investment funds should take the lead in being patient capital, give play to the pulling role of the National Emerging Industry Venture Investment Guidance Fund and the National SME Development Fund, ensure that funds tilt toward the foundational layers of AI and early-stage projects, do not compete with the market for profits, and do not chase short-term book returns. Third, cultivate angel investors and early-stage investment institutions, support the establishment of angel guidance funds in places where universities and research institutes are concentrated, and encourage entrepreneurs with industry experience and technical experts to participate in early-stage investment, so that the money with the best vision enters at the earliest stage. When the source has long-term money, the growth of AI enterprises can receive long-term and strong accompaniment. Improve incentive mechanisms. With fault tolerance and incentives as the key, enable all types of capital to dare to invest early and be willing to invest long term. Having long-term money, we must also solve the problem of “not daring to invest.” Early-stage AI projects inherently have a high failure rate; it is the industry norm for two or three out of ten projects to succeed. However, traditional venture capital regularly assesses book profits and losses, and investment losses must face audit accountability. Winning is considered one’s duty, while losing requires bearing responsibility, so management teams naturally tend to cluster in later stages and compete for highly certain projects, while the seed stage that truly needs timely help receives little attention. To make capital dare to take the risks it should take, institutions must be relaxed and incentives must be in place. First, reform the assessment methods for state-owned venture capital, implement long-cycle and phased assessment in accordance with the laws of scientific and technological innovation, lengthen the assessment window, reduce the weight of annual profit indicators, put the long-term effects of cultivating hard-tech enterprises and driving industrial upgrading to the fore, and allow early-stage investment portfolios to offset losses with gains and be accounted for as a whole. Second, implement due diligence, compliance, and exemption from liability, clarify the boundary between diligence and dereliction of duty, and specify the standards for identifying normal investment losses and exemption procedures. As long as decision-making procedures are compliant and there is no transfer of benefits, individuals must not be held responsible for project failures, so that those who invest early and in small firms can unload their burdens. Third, improve risk-sharing and positive incentive mechanisms, make good use of science and technology innovation risk compensation funds, promote the practice of government, banks, guarantees, and funds jointly bearing risk, and allow government guidance funds to transfer part of their returns to social capital. Implement market-based compensation and performance fees for fund managers, so that the long-term efforts of professional teams receive reasonable returns. We must strive to open up space for bold experimentation and trial and error, so as to provide a broader stage for patient capital. Clear the exit channels. With smooth exits as the hub, form a capital cycle in which capital can be invested, exited, and reinvested. Whether patient capital can exit smoothly and obtain reasonable returns determines whether long-term money dares to come and is willing to stay. In reality, many funds are “trapped” by projects; AI companies generally take about ten years from founding to listing, and whenever the IPO pace fluctuates, maturing funds can only tough it out. The M&A market is also not active enough; there are few transactions in which large companies acquire startups, private equity transfer platforms are still being cultivated, share pricing is difficult, there are few buyers, and a single exit channel has led capital to crowd into a small number of pre-IPO projects. To open up the cycle, more doors must be opened. We should make good use of the institutional inclusiveness of the STAR Market, ChiNext, and the Beijing Stock Exchange, stabilize expectations for the listing of unprofitable hard-tech companies, and let AI companies that truly master key technologies see a path to the capital market. We should place M&A and restructuring in a more prominent position, encourage industry leaders to integrate technology, teams, and computing power resources through M&A, and shorten the innovation chain through “acquisition-based R&D.” We should optimize review arrangements and reduce transaction costs to provide venture capital with normalized exit channels. We should also cultivate the private equity secondary market, expand pilot programs for share transfers, straighten out valuation and pricing methods, and develop continuation funds and S funds, so that shares of maturing funds have someone to take them over and prices can be agreed upon. We should fully leverage the incubation and circulation functions of regional equity markets to preserve financing and transfer channels for small and medium-sized enterprises that do not yet meet listing conditions. Investment, cultivation, exit, and reinvestment are closely linked; if the exit end is opened up by one inch, the early-stage end will become active by one foot, and only then can the capital cycle in the AI field truly turn. Competition in artificial intelligence appears on the surface to be a contest of technology and products, but behind it is a contest of capital endurance and institutional quality. Cultivating patient capital requires a complete set of institutional arrangements that make long-term money dare to come, dare to invest, able to exit, and willing to reinvest—from broadening funding sources to improving fault-tolerant incentives, and then to clearing exit cycles—all linked and indispensable. Putting this set of arrangements into practice and guiding more capital to walk in the same direction as AI companies and achieve mutual success will smooth the virtuous cycle of technology, industry, and finance, enable the steed of artificial intelligence to run faster and more steadily, and accumulate enduring momentum for developing new quality productive forces and gaining the initiative in international competition. (Author affiliation: Shanghai University of Finance and Economics)