news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

as a group of companies with strong innovation capabilities and huge growth potential, the number and activity of unicorn companies are important indicators of a country's and region's innovation capabilities and innovation ecology, and are also important innovation entities for enhancing international and regional competitiveness.

text | liu yiqin, reporter of caijing

editor | xie lirong

in march 2024, 70 employees of inflection, an american ai startup, met with microsoft ceo satya nadella in a hotel auditorium. the startup has caught up with the big model trend, raised $1.5 billion in the past two years, and is valued at $4 billion, making it the third-largest big model unicorn in silicon valley. now, they need to consider whether they are willing to join microsoft after the company's core technology is acquired by microsoft.

an employee asked if he would become conservative if he joined microsoft because the company is too big and developing slowly. nadella asked, "are you willing to change a $3 trillion company?"

a long-time silicon valley investor said nadella’s daily job is to manage “a digital kingdom,” but he will still be deeply involved in the specific matters of investing in large ai models.

american technology giants are keen to compete for potential technology companies. since the beginning of this year, there have been more and more similar acquisitions in silicon valley. in august, google acquired character.ai for $2.5 billion; in july, amazon acquired adept ai's core technology and most of its employees.

silicon valley in the united states is the leader of this round of ai innovation and entrepreneurship. as of now, there are 27 unicorn companies related to big models in silicon valley (referring to unlisted companies established less than ten years ago and valued at more than $1 billion). china is closely behind, and five new unicorn companies in the field of ai big models have emerged, namely zhipu ai, baichuan intelligence, dark side of the moon, minimax, and zero one everything. among them, zhipu ai has the highest valuation, about 20 billion yuan (about 3 billion us dollars); zero one everything has the lowest valuation, about 7 billion yuan (about 1 billion us dollars).

europe also has several unicorn companies related to big models, but the number is generally small, which is consistent with the previous trend. they include french ai data management company dataiku (valued at $3.7 billion) and basic big model company mistral ai (valued at $6 billion); british ai data company quantexa (valued at $1.8 billion) and text and graphics company stability ai (valued at $4 billion).

in the last round of ai innovation cycle led by deep learning, china and the united states were the two poles of the world. in this round of innovation cycle led by the aigc big model, china and the united states are still the two fastest countries, but there are obvious differences.

in silicon valley, startups in the field of big models have formed a relatively complete ecosystem: there are leading startups represented by openai, and there are also competitors that can match them; there are application companies based on big models, which can also quickly obtain users and financing; there are a large number of tool-based startups that serve big models, and their valuations have been rising; platform giants are deeply involved, they have their own models, open ecosystems to startups, and provide financing and acquisition opportunities. the total amount of financing in the ai ​​field in the united states in 2023 will exceed us$50 billion.

in china, the new round of technological revolution cycle has limited stimulation for venture capital enthusiasm. according to data from it orange, a venture capital data service provider, financing in china's ai field has been rising since 2014, with a total financing amount of approximately 441.1 billion yuan in 2021. it will drop to 157.9 billion yuan in 2022, a decrease of 64%. in 2023, with the popularity of large models, it will still show a downward trend, with annual financing of approximately 110.1 billion yuan. china's top startups are all based on basic large models; giant companies represented by alibaba, tencent, and meituan are actively investing, but competition is greater than cooperation; the application and tool links are relatively weak, and there are many related companies, but few of them are large-scale.

according to people’s daily, on may 23 this year, at a symposium for enterprises and experts held in jinan, shandong province, xi jinping, general secretary of the communist party of china, president of china, and chairman of the central military commission, asked after hearing a speech on innovation and investment: “what is the main reason for the decline in the number of new unicorn companies?”

on april 30 this year, the political bureau of the cpc central committee held a meeting chaired by general secretary xi jinping to analyze and study the current economic situation and economic work. the meeting emphasized the need to actively develop venture capital and strengthen patient capital.

the so-called "patient capital" refers to capital that does not pursue short-term returns as its primary goal, but focuses on long-term projects or investment activities and has a high tolerance for risk. the "patience" of patient capital emphasizes equity investment in the early stages of a company's growth, rather than stock investment after the company is successfully listed. its core characteristics are early investment, small investment, long-term investment, and investment in hard technology. when the company grows and develops, the mission of patient capital has been completed, and it can gradually exit with sufficient investment returns.

the political bureau of the cpc central committee held a meeting on july 30 and pointed out that it is necessary to effectively support the development of gazelle enterprises and unicorn enterprises. in nature, gazelles are known for their agility, small size, fast running and high jumping. gazelle enterprises have strong innovation capabilities, new professional fields and great development potential. they mainly include small and medium-sized enterprises that have entered a high growth period supported by technological innovation or business model innovation.

unicorn companies generally refer to unlisted companies that have been established for no more than ten years, have a valuation of more than us$1 billion (a small number have a valuation of more than us$10 billion), and have unique core technologies, unique competitive advantages and market potential.

gazelle companies are generally considered to be the reserve team of unicorn companies. unicorn companies are often characterized by high valuations, start-ups, and business models that are difficult to replicate. the companies themselves often have a significant impact on their industries.

from international experience, the number of gazelle enterprises in a country reflects the innovation capability and development speed of the region to a certain extent. as a group of enterprises with strong innovation capability and huge growth potential, the number and activity of unicorn enterprises are important indicators of the innovation capability and innovation ecology of a country and region, and are also important innovation entities to enhance international and regional competitiveness.

many entrepreneurs and investors interviewed by caijing believe that the current entrepreneurial ecosystem in china's ai field is not yet sufficient to support the development of more gazelle and unicorn companies. but where exactly is it immature, and how should it be improved? to answer this question, we need to gradually disassemble and compare, layer by layer.

the current growth status of ai unicorns

in this round of ai startups in silicon valley, in addition to the underlying model innovation represented by openai, there are also many companies related to large model applications and large model tools, which have also received high valuations.

driven by microsoft and openai, the global ai industry has developed rapidly in the past two years. at present, there are 37 generative ai-related unicorn companies in the world, with 17 new ones added in the past year. among them, 27 are from the united states and 5 are from china. the five chinese companies are: zhipu ai, baichuan intelligence, dark side of the moon, minimax, and zero one everything. zhipu, china's most valuable ai unicorn company, has a pre-investment valuation of approximately 20 billion yuan (approximately 3 billion us dollars) in the latest round of financing. the post-investment valuation of the latest round of financing on september 5 this year has not yet been disclosed. since its establishment in 2009, zhipu has completed 11 rounds of financing.

the estimated valuation of openai's latest round of financing will reach $100 billion. if the financing goes smoothly, openai will become the world's second largest unicorn company, second only to bytedance. openai's direct competitor anthropic is currently valued at $18.4 billion. in march this year, amazon invested an additional $2.75 billion in the company, which is the largest venture capital investment in amazon's history. although the valuation gap between the two companies is obvious, anthropic has only been established for three years, while openai has been established for nearly ten years.

minimax, a chinese ai unicorn company founded in 2021 like anthropic, is currently valued at approximately $2.5 billion. among the five large model unicorns in china, zhipu ai, baichuan intelligence, and dark side of the moon are all valued at approximately $3 billion, and zero one everything is valued at approximately $1 billion.

the main businesses of the five chinese unicorn companies are all basic large models, and their current scale is relatively small. the combined valuation of the five companies is less than that of anthropic alone.

currently, silicon valley has inspired the emergence of more unicorn companies beyond basic large models. midjourney, an ai painting tool company founded in 2021, was valued at $10 billion last year with annual revenue of $200 million. painting tools were also one of the first large model application companies to become popular.

glean, an enterprise search service company founded in 2019, was valued at $4.5 billion in its latest round of financing, double the amount six months ago. the company's business does not sound complicated. it mainly helps corporate employees search for internal information, which is an enterprise office software based on a large model. according to media reports, glean's revenue in 2023 is about $39 million, and its revenue in the first seven months of 2024 is about $55 million.

between the underlying basic big model and the upper-level application, there is another "rich mine", which is the tools and service providers of the middle-level big model. many investors told caixin that this is the layer that silicon valley is best at and most mature, and it is also a relatively weak link in china's technology field.

on september 6, 2024, some of the big model technologies displayed at the shanghai 2024 bund conference: ant group (bai ling native multimodal big model), shengshu (vidu video big model), baichuan intelligence (baichuan big model), etc. photo/ic

ai development is inseparable from data. before data can be used for training, it needs to be processed and labeled. data labeling is the "dirty and tiring work" in the field of ai. before the last round of ai boom, china already had a large number of companies related to data labeling, but few of them were large-scale. many chinese technology companies can only choose to set up their own data labeling teams to ensure data quality.

founded in 2016, scale ai, an american data annotation company, has a latest valuation of $13.8 billion. the company's revenue in 2023 is about $330 million, with a gross profit margin of about 53%. it is expected to have revenue of about $1 billion in 2024. almost all of the top american technology companies are its customers. while most ai startups are still suffering from huge losses, scale ai is close to breaking even.

in addition to the newly added unicorns, there are also a large number of former unicorns in the united states that have transformed into large-scale models. the most representative of these is databricks, which has been established for more than ten years.

databricks was founded in 2013 by a team of professors and phds from the university of california, berkeley. its main business in the early days was cloud computing-based data analysis services. the company is a technology unicorn with a valuation second only to bytedance and openai among currently unlisted companies. in 2023, the company's revenue was $1.6 billion, and its latest valuation was $43 billion.

it did not build its own cloud platform. it first cooperated with amazon aws and then started to cooperate with microsoft azure in 2016.

after openai became popular, many people in the industry believed that previous generation ai companies like databricks would be replaced. at the beginning, the company's ceo (chief executive officer) ali ghodsi's external speeches were indeed more or less confrontational with the emergence of new technologies. he repeatedly declared that large models are not omnipotent, and using large models to replace existing data processing tools will be very costly and inaccurate. however, in january of this year, databricks quickly launched its own large model, which is based on meta's open source llm model. its founder said that the biggest feature of this model is that the cost is lower than gpt-3.5, and it can work with its own big data open source software, which can give customers more choices and help the industry reduce the cost of large models.

china's ai startup ecosystem is different. the five leading unicorn companies are all concentrated in the field of basic large models, and have not yet spread to other circles. the leading technology companies are indeed vigorously deploying large models. most of them have taken charge of everything from basic large models to upper-level applications. this has, to a certain extent, prevented everyone's valuation from increasing rapidly.

the information conveyed by many investors surveyed by caijing is that at this stage, the valuation of basic large models in the chinese capital market will be higher, mainly based on three reasons: first, the technology is still being updated and iterated. if applications are based solely on open source models, they are easily covered by new technologies; on the other hand, the current technology maturity has not yet reached the stage where it can be applied on a large scale. in addition, due to high costs, the b-end application of large models will precede the c-end application, and it is difficult for b-end applications in the current chinese market to start in volume.

in china's large-scale model entrepreneurship, the weak middle layer is also due to the lack of willingness of b-side users to pay. at the same time, if chinese corporate users have tool needs, they will naturally give priority to foreign products and services.

where do the differences between china and the united states come from?

the business models, user composition and investment logic of different markets jointly determine the valuation and development direction of this round of ai startups.

at present, the most concerned issues of ai big models in the venture capital circles of china and the united states are different. chinese investors have begun to pay attention to the commercialization of ai unicorns, because if they cannot prove their commercialization capabilities, it will be difficult to continue to obtain financing and maintain a high valuation.

in silicon valley, technology companies and investors are mostly discussing in public how to make further technological breakthroughs and how to ensure ai security; in private, they often talk about the fact that the data for training large models is almost exhausted.

in may this year, openai ceo altman mentioned at a technical conference that ai companies will soon exhaust all available data on the internet. a person who has surveyed many american technology companies also mentioned that insufficient data is the most mentioned problem.

the quality of large ai models mainly depends on three factors: model architecture, computing power, and data. at present, the industry generally recognizes that transformer is the mainstream architecture and will not be overturned in the short term. computing power depends on performance and scale, the larger the better. data is relatively complex and requires data diversity, data quality, and data quantity.

to some extent, data quality and commercialization capabilities are inseparable. qiu zhun, an overseas partner of huaying capital, believes that in the era of big models, data is a "threshold". the main reason why there are relatively few big model companies in china that have real application capabilities is the lack of data capabilities. how to process data and how to combine data with algorithms have strong technical requirements and are very difficult.

silicon valley unicorns have not had smooth commercialization. mustafa suleiman, the founder of inflection who joined microsoft, once publicly stated that the company's business model was unsuccessful. according to media reports, the gross profit margin of large model companies including openai and anthropic is about 50%, and the cost of model training is not included.

maintaining such a gross profit margin means that it is difficult for the company to make a profit. according to statistics from meritech capital, a us venture capital firm, the average gross profit margin of listed software companies is 77%. the gross profit margins of well-known software companies such as gitlab, adobe, and uipath are all around 90%.

the reason why the gross profit margins of large-scale unicorns in the united states are relatively low is not only that various costs remain high, but also that many of their users come from large cloud platforms, and cloud vendors will take commissions from them, which reduces the gross profit.

the calculation method used by analog software companies is not universal in the chinese market. chinese ai companies have always had their own unique business model - doing customized projects.

according to data from third-party data agency idc, the market size of china's large model platform and related applications will reach 1.765 billion yuan in 2023. the top three in market share are baidu, sensetime and zhipu ai.

according to multiple public information platforms, in the first half of 2024, china's large-scale model bidding announcements exceeded 230, with a total amount of more than 1 billion yuan disclosed (about 30% of the projects did not disclose the amount). the customers were mainly state-owned enterprises and the government, and the winning manufacturers were also mainly state-owned enterprises, technology giants and local system integrators, including china telecom, china mobile, iflytek, alibaba cloud, tencent cloud, huawei cloud, etc. among the five large-scale model unicorn companies, only zhipu ai has won multiple bids.

among the publicly awarded projects, computing power and customized applications are the absolute main force, and most of the large orders are for computing power services, which does not provide obvious competitiveness for entrepreneurial companies.

idc, a third-party market analysis agency, mentioned that in 2023, the domestic industry will make more early investments in big models, or even wait and see without heavy investment, so the overall market size in 2023 will not be significant. in 2024, leading internet companies will increase their investment in big models and launch price wars, bringing certain competitive pressure to early big model startups.

at the beginning of this year, a chinese large-scale model unicorn company received an order worth tens of millions of yuan from a local government. the investor of this unicorn company told caijing that currently orders of similar scale in the industry are all from government customers. in the past few years, local governments have done a lot of smart city projects, so there are many related service providers. after the government purchases large models, it distributes them to service providers to build an ecosystem together.

different business models also mean different valuation methods. the valuation of a startup is affected by multiple factors, which are ultimately reflected in the business model, but the business model itself is closely related to the characteristics of the market. in the united states, investors are very familiar with the development path of software companies. startups can focus on polishing their products and improving their technical level. if the product is good enough, there will naturally be user growth and a scale effect.

in silicon valley, most of the large-scale unicorns are currently in a loss-making state, and the capital market is willing to give high valuations. in addition to the establishment of a business model, there are relatively wide exit channels. mergers and acquisitions and ipos (initial public offerings) are the main exit channels. in silicon valley, most technology startups are acquired through mergers and acquisitions. today, in order to stabilize their position as soon as possible, the pace of mergers and acquisitions by the giant american technology companies has significantly accelerated.

in china, many investors have mentioned that if a startup claims to be making software, it is very likely that it will not be able to get financing. entrepreneurs will be repeatedly questioned, if it is customized software, how to polish and iterate the product? how to increase the volume? can they break through the "bottleneck" technology?

qiu zhun mentioned that many ai companies currently serve "key users" and need to provide on-site delivery, which means that these ai service companies may become labor-intensive companies. "if all your capabilities are in delivery, then you are definitely not a software company. the capabilities of a software company should be in r&d and products, and delivery is only a part of it."

for chinese investors, compared with data such as the number of customers and order volume, they pay more attention to whether a startup has a strong enough background, including the background of the company's core members and the background of its customers. for example, whether it can get orders from first-tier city governments or large state-owned enterprises, whether there are well-known experts or former executives of large companies among the company's executives, etc.

“it’s not that we don’t recognize standardized products, but the reality of the chinese market forces us to work in this direction.” an investor mentioned, “this may be a unique entrepreneurial ecosystem in china.”

a chinese ai startup started working on a large model framework in 2017. the company's founder is committed to making standardized ai tools and has accumulated some well-known corporate users and a large number of developer users. the company is currently facing a tight cash flow. the founder mentioned that the r&d, testing and optimization stages require high costs, and he could not get enough financing; he tried to form a new team to do customized projects to make money to support the company, but soon found that it also takes all his efforts to do this well. once the focus is dispersed, the product will lag behind foreign open source tools.

the scale of financing, valuation, entrepreneurial direction, etc. are just a small part of the iceberg. what can support these is the larger entrepreneurial ecosystem beneath the water.

the importance of ecological openness

the ecosystem of silicon valley has also undergone tremendous changes in the past two years. the main investors have changed from investment institutions to technology giants, but the open ecological environment has been retained.

before 2023, scale ai, a us data annotation company, was in a state of long-term losses, and its revenue growth was also lower than expected - revenue growth in 2022 was less than 50%. it is a typical company that benefits from the explosion of the large model industry. the company began to cooperate with openai in 2019. last year, the two companies discussed a merger, but nothing came of it.

on june 10, 2024, at apple headquarters in california, openai ceo sam altman (center) attended the worldwide developers conference (wwdc). photo/ic

when other companies began to catch up with openai, they naturally became scale's customers, which led to rapid growth in the company's revenue. this is similar to the growth path of apple's supply chain companies. when an iconic leading company emerges in an industry, it will drive other links in the industry chain to rise.

xu chenyang was the general manager of siemens innovation center. he has been an investor in silicon valley for eight years and has seen more than 3,000 startups. he has also worked in europe, china, and the eastern united states. he feels that silicon valley still has the best entrepreneurial ecosystem in the world.

entrepreneurial ecology sounds like a broad description. xu chenyang gave an example. if a gas station is doing very well, other people who want to make money will open small shops, restaurants, hotels, etc. near the gas station, and gradually a small village will be formed here. "this is the ecology." in other markets, if a gas station becomes popular, a large number of gas stations will soon appear around it. in the end, everyone's business is not good, and it cannot grow into an "ecosystem."

one of the characteristics of silicon valley's entrepreneurial ecosystem is openness. xu chenyang once held a party in silicon valley with more than 20 participants, mainly engineers and executives from technology companies such as apple, google, netflix, and meta. everyone talked about current technology trends, what they are currently doing, and what new technology directions are worth paying attention to. almost anything can be discussed. one of the few "outliers" in silicon valley is apple, and nothing can be discussed about apple.

a few years later, xu chenyang returned to china to start a business and often participated in some ai forums and conferences. he felt that although everyone communicated frequently, the value was not great. what everyone talked about was either self-promotional content or some public information.

silicon valley's openness is also reflected in the flow of talent. california had already enacted law to ban non-compete agreements in 1872, and a large number of ai companies in silicon valley have talent from technology giants.

in the early years, the chinese venture capital circle was talking about the silicon valley garage startup culture. a few young people started a business together. because they had no money, they worked in the garage. the garage owner did not charge rent, but only a small amount of shares. it seemed that silicon valley was full of angel investors.

what is behind this is the entrepreneurial services of silicon valley. technology has been integrated into silicon valley. in silicon valley, almost all companies claim to be technology companies. financial companies need to add technology, law firms need to add technology, consulting agencies need to add technology, and even human resources need to add technology. these "technologies" largely refer to their growing together with silicon valley's technology companies. a technology startup can use shares to exchange for these corporate services.

in addition, talent is the core of the innovation ecosystem. most of the earliest talents in silicon valley came from stanford university, and the first 100 employees of google all graduated from stanford. stanford university and another famous school in silicon valley, the university of california, berkeley, both have a large number of courses designed specifically for cultivating entrepreneurs, and their alumni are spread across the fields of technology and investment.

in addition to local talent training, silicon valley startups and giant technology companies also have a large number of people from all over the world. the basis of today's large model is the transformer architecture proposed in the paper "attention is all you need" co-authored by eight google scientists in 2017. these eight scientists come from eight different countries.

on september 4, 2024, the iflytek ai learning machine artificial intelligence technology experience exhibition area at the 2024 china international electromechanical products expo and wuhan international industrial expo. photo/ic

when xu chenyang first arrived in silicon valley more than a decade ago, silicon valley's technology entrepreneurs were not very willing to accept investment from large companies. the main reason was that investment decisions of large companies were relatively complex. from the initial contact to the confirmation of investment, a relatively long period was required. the person in charge of investment needed to report to the group and discuss with the business department whether there was business synergy, etc. in investment institutions, in many cases, only the consent of the partners was required to make the payment immediately.

moreover, after receiving investment from a giant, there will usually be business cooperation, and the cooperation process will be very slow. xu chenyang said that there were many such examples in silicon valley in the past, and many startups failed to grow and died because of this. giants usually don’t care so much about the return on investment. for example, investing $100 million and getting a return of $1 billion is a very successful investment for ordinary investment institutions, but $1 billion is just a small amount in the overall income of the giant.

over the past decade, silicon valley giants have made adjustments to their outbound investment models, including allowing investment departments to make independent decisions and optimizing processes. these adjustments have streamlined their mechanisms for bringing potential unicorns under their control, allowing technology giants to become an important force in technology venture capital; on the other hand, they have also given startups more opportunities to grow into unicorns.

the impact of complex venture capital relationships

when venture capital becomes a technology giant, the relationship between the two parties becomes more complicated.

the three most important roles in the entrepreneurial ecosystem are entrepreneurs, investors, and entrepreneurs. ideally, these three types of people perform their respective duties. entrepreneurs are responsible for innovation and breaking rules; entrepreneurs provide platforms, open cooperation, and expand the ecosystem; investors provide funds to help entrepreneurs get through the start-up and rapid development periods and get returns. in a good entrepreneurial ecosystem, entrepreneurs are the protagonists, or in other words, only entrepreneurs who can be the protagonists can make real innovations.

in this round of big model startup boom in the united states, the struggle between the leaders is concentrated in the "palace fight" segment of openai.

before november 2022, almost no one expected that the global technology industry would usher in a new wave of shocks, including openai itself. in mid-november of that year, altman and several executives met to discuss how to solve the technical problems of gpt-4. during the meeting, altman suddenly decided that he would release chatgpt, which has weaker functions. at that time, openai employees thought that this would be a piece of news that no one would pay attention to.

in fact, before this, in september 2022, meta had released an ai chatbot that could help users write articles and do math problems. it was shut down three days after it went online, and meta said it needed to adjust its product strategy. at that time, meta's focus was still on virtual reality.

on november 30, 2022, chatgpt was launched. due to too many registrations and lack of advance preparation, openai's company server crashed almost instantly. five days after its launch, the number of users exceeded 1 million, and a few weeks later, the number of users exceeded 100 million.

silicon valley tech giants such as google, meta, amazon, and xai have all felt the drastic impact, except microsoft. according to media reports, two months before chatgpt went online, nadella showed gpt-4 to microsoft executives, and he asked the entire microsoft to make a strategic shift around this technology.

since then, openai and its founder altman have become the most watched stars in the global technology field. technological breakthroughs are one aspect, and the company's series of dramatic developments are also eye-catching. last year, altman was once dismissed by the company's board of directors, which staged a "palace drama". nadella was deeply involved and responded to and solved various problems in the first time.

as things have developed to this point, nadella, as an investor and platform owner, has become another protagonist in this story. and his attitude is subtly changing.

according to media reports in august 2023, microsoft plans to launch a new cooperation with databricks to sell its software on the azure platform, which can help users develop their own large models from scratch or better use open source large models. not only that, microsoft also created a chatbot through openai's technology to specifically serve users who are not very tech-savvy to use databricks' software.

the industry believes that microsoft's move is to prevent its users from over-relying on openai. this judgment is basically correct, but microsoft has done more than just the above actions.

many people are paying attention to the details of openai's "palace fight". in fact, the trend of the incident is closely centered on the competition for dominance between technology giants and entrepreneurs. in november 2023, as the largest supporter and beneficiary of openai, microsoft began to worry about the risks of relying on a single model in advance.

the means to mitigate the risk are actually very conventional - spread the eggs in different baskets. microsoft quickly formed a new team to develop large models with smaller scale and lower operating costs. in addition, microsoft gave more traffic to other large model products on its cloud platform azure, including models of openai's competitors, canadian ai unicorn cohere and french ai unicorn mistral.ai.

in january 2024, nadella stated at the davos forum that developing small ai models is a way to "control our own destiny" and that "in the future we will have a variety of capabilities and models."

after that, microsoft obtained the technology license of inflection and recruited most of the company's employees to help microsoft add new weight to the research and development of its own models.

the relationship between giant companies and startups is always complicated. because of their strong strength, large technology companies usually have multiple identities as investors, platform providers, and competitors. they are often both "referees" and "athletes."

in china, whether or not a startup should accept investment from a giant company is often a difficult question to answer. ten years ago, most chinese entrepreneurs were happy to accept it, believing that the giants were not only generous in terms of valuation, but also could bring in resources and customers, killing two birds with one stone. but soon, entrepreneurs discovered the contradiction. the giants were not philanthropists. in their eyes, investing in startups was a new "fuel" for their own business development, or a means to prevent future competition in advance.

whether in china or the united states, it is not a cost-effective business for giant companies to acquire a startup with only dozens of people at a high price. an investor of a chinese technology giant told caijing that in addition to industry layout, they also consider investing in large model companies so that they can observe the progress of these startups up close, "if there is value, they can follow up as soon as possible."

due to strict antitrust and technology protection, american technology giants usually have to pay high prices to acquire startups. a startup that xu chenyang once invested in was facing bankruptcy due to poor development, but it was still sold to meta at a high price because the company had a core technology that meta believed might be used in the future.

the valuation of large-model startups is growing very fast. microsoft has studied and practiced new acquisition models, acquiring the core technology and team of inflection, which was valued at $4 billion, for only $650 million. google and amazon also used the same approach to acquire two other large-model unicorn companies.

whether in china or the united states, cooperation between startups and giant companies is important and dangerous. it is important that entrepreneurs understand the benefits of cooperation and the dark side. entrepreneurs in china and the united states have also summed up similar experiences - cooperation with multiple giant companies can effectively reduce the risk.

according to media reports, in addition to microsoft, nvidia and apple have also participated in the discussion of openai's recent new round of financing.

current focus

the biggest difference between silicon valley and china’s innovation ecosystems is the number of cycles they’ve gone through.

silicon valley’s entrepreneurial ecosystem in the field of technological innovation is better than china’s, but that doesn’t mean silicon valley is always in the lead.

silicon valley has also experienced a "dark period" in venture financing. around 2000, silicon valley fell into the "internet bubble" crisis. venture capital increased from $8 billion in 1990 to $100 billion in 2000. in 1999, 55% of venture capital projects were internet projects, and more than 150 internet projects were listed in 1999. from 2000 to 2002, about 1,000 internet companies went bankrupt and more than 3,800 were merged. during that period, almost all new investments disappeared.

china's venture capital market has been cooling since 2019, and then experienced the epidemic period and the withdrawal of foreign capital. according to data from third-party data agency enterprise business card, china's primary market completed a total of 861.7 billion yuan in financing in 2022, a year-on-year decrease of 38%, and continued to decline by 13% in 2023. "compared with the dark period of silicon valley, china's market is not bad," an investor told caixin, "the biggest difference is the number of cycles it has experienced."

if we count from the founding of stanford university in 1891, silicon valley has been the global center of technological innovation for more than 100 years. in the middle of the last century, many technology companies mainly engaged in semiconductors were born here. after that, many technological innovation companies such as intel, amd, apple, oracle, cisco, google, and tesla have grown and developed here. after these giant companies went public, they created high returns for investors, and there is a large amount of capital in the market to explore new technological projects.

in china, the first market-oriented venture capital institution to enter china was idg capital of the united states. in 1989, idg capital invested in its first project in china. since then, a large number of us dollar funds have entered the chinese market, stimulating the rapid development of china's internet industry to a certain extent. after 2000, china's venture capital field has been almost one hot spot after another, from the internet to mobile internet, to "internet +", "mass entrepreneurship and innovation" and "hard technology". until 2019, a large number of investment institutions encountered difficulties in fundraising and entered a short trough period.

today, outside of the ai ​​field, china and the united states are still the two major centers of global technological innovation and entrepreneurship, with huge advantages over other countries. there are currently about 2,000 unicorn companies in the world, with 714 in the united states and 675 in china, and the gap between the two is not large. india, which ranks third, has 86 unicorn companies, and the united kingdom, which ranks fourth, has 64.

having more unicorn companies means that a region has stronger innovation capabilities and better development prospects, but incubating and nurturing such companies is not easy and requires a better ecological environment, especially the investment and financing environment.

after 2022, china's primary market has undergone a structural change: dominated by state-owned capital, affected by policies and geopolitics, the investment theme has changed from scale, income, and profit to "specialization and innovation", "supply chain security", "new quality productivity", "solving the 'bottleneck' problem", etc., and the investment direction is concentrated in semiconductors, new materials, intelligent manufacturing, new energy, aerospace and other fields. in china, market orientation and policy orientation are almost the same, and policy makers are also market participants.

in the united states, the main responsibilities for the ai ​​ecosystem are borne by giant companies such as microsoft, google, and amazon, which provide funds, platforms, application scenarios, and customers. li jiaqing, president of legend capital, told caixin that in china, the government, especially the beijing and shanghai governments, bears similar responsibilities. these are two different models, determined by different economic systems and development models. it is difficult to say which is better, "but at this stage, we must admit that the united states is ahead."

xu chenyang found that before 2022, when china's venture capital was developing rapidly, the investment styles of investors in china and the united states for technology companies were very different. american investors not only paid attention to the technical capabilities of startups, but also paid more attention to business models and specific execution teams, and would spend a lot of energy to carefully study all aspects of the company.

many chinese investors have switched from investing in the internet industry to investing in technology. in many cases, they pay special attention to the background of the founder, such as whether the founder is an academician or has returned from a prestigious overseas university. some investors even directly say, "if you can recruit an academician, i will invest."

but in silicon valley, even if the three founders of a company are nobel prize winners, investors will investigate and study the company thoroughly before deciding whether to invest and how to value it.

in the past two years, the enthusiasm for financing in china's primary market has declined, and the overall scale of investment has decreased. instead, investors will observe a technology startup more carefully.

investors learn and adapt to new models in the cycle, and so do entrepreneurs. silicon valley investors believe that most excellent entrepreneurs have the ability to go from 0 to 1, while a company from 1 to 10, from 10 to 100, requires managers with different abilities at different stages. just like teachers are divided into elementary school, junior high school, high school, and college teachers, it is almost impossible for a teacher to lead a student from elementary school to college graduation.

therefore, in silicon valley, when startups reach a certain stage of development, they are frequently acquired by large companies or undergo a major change in senior management. over time, silicon valley has seen the emergence of many entrepreneurial talents with comprehensive strengths - those who have had multiple entrepreneurial experiences and are prepared for the problems that may be encountered in the early stages of entrepreneurship; some people who have sold their startups will turn to investment or join the acquirer, and in this process they will gradually acquire the perspective of investment and large companies.

another important participant in the entrepreneurial ecosystem, large companies, also need a longer time to grow. all large companies recognize that one company cannot do everything, and the ecosystem needs to be built together. only with more participants can the platform be vibrant. a person from a large company told caixin that china's large companies generally have a sense of anxiety and can't help but "roll" and often have the urge to do it themselves. this is determined by market reality. on the one hand, china's software ecosystem itself is not rich enough, and large companies need to play a leading role; on the other hand, if their own platform size is placed in the global market, it is not large enough and the confidence is not that strong.

it is difficult for people to have "patience" when they are very young, and they always need immediate feedback. it is difficult for a company that has been established for ten years to make real long-term plans. based on the current situation, to cultivate more unicorn companies, government guidance, diversified investment, market operation, and the injection of more and more long-term capital and patient capital are the only way.

where will more long-term and patient capital come from? it is imperative to accelerate the construction of a science and technology financial service system that is highly compatible with the development of innovative industries, and to open up channels that convert social funds into patient capital and effectively flow it to science and technology enterprises.