"Financing, investment flow, and self-sustainability: the 'rising stars' of large models race to monetize!"

09/30 2024 472

By Wang Huiying

Edited by Ziye

Halfway through 2024, hailed as the "Year of AI Large Model Applications," the industry has undergone rapid changes amidst various voices.

The once-lengthy technological development cycle has been accelerated by large models. In just two years, players have engaged in the "Hundred Models War," and capital has poured in, fueling exploration amidst skepticism.

As we enter the second half of the year, the industry's direction becomes more nuanced.

Latest data from Similarweb indicates a steep decline in monthly visits to ChatGPT's website; OpenAI's product release pace has slowed; NVIDIA's share price has fluctuated, dropping below its intraday high in the US stock market; and domestic leaders have also been critical of large models...

On the other hand, domestic "rising stars" of large models remain hot. MoonShadow AI was reportedly invested in by Tencent, with a post-investment valuation of $3 billion; Zhipu AI secured a new round of financing worth billions of yuan, with a pre-investment valuation of 20 billion yuan; and Fei-Fei Li's maiden venture, World Labs, announced a $230 million funding round, backed by NVIDIA.

The industry finds itself in a state of "fire and ice," as the market gradually dispels the myth of large models, and capital returns to rationality. Only when the hype subsides will truly valuable enterprises or applications emerge.

Regardless of the stage, commercialization is a recurring topic in the large model industry. Large models are unique in their need for "burning money," with significant costs for technological research and development and operational expenses. This poses a tight constraint for large model enterprises.

Especially for startups like MoonShadow AI and Zhipu AI, lacking the financial resources and abundance of giants, relying solely on financing for blood transfusions without self-sustainability will inevitably lead to elimination.

The shifting tides propel the large model industry forward. Moving from the "Hundred Models War" to the "Application War," the industry is poised to enter a new phase where finding a unique monetization path is crucial, in addition to products and applications.

1. "Star Enterprises" Remain Hot

During the recent 2024 Cloud Town Conference keynote, Alibaba CEO Wu Yongming made an apt analogy: last year, large models' mathematical abilities were akin to middle school students, but today, they can compete for international math olympiad gold medals, particularly in subjects like physics, chemistry, and biology, approaching doctoral levels.

In the past 22 months, the advent of large models has accelerated AI development unprecedentedly. From the Hundred Models War to the Application War, internet giants have joined the fray, and numerous AI-related startups have emerged.

After several rounds of elimination, China has given birth to numerous star enterprises, from the "New AI Four Dragons" to the "AI Five Tigers" and "AI Six Strong," all fueled by capital.

CBInsights data shows that generative AI startups globally secured approximately $20.4 billion in financing in 2023, over five times the $3.6 billion raised in 2022.

The hype continues.

In August, ZeroOne received a new round of financing worth hundreds of millions of dollars, with a latest valuation of 10.4 billion yuan. The same month, MoonShadow AI also secured a $300 million funding round. In July, Bichuan AI completed an A-round financing of 5 billion yuan and launched a B-round financing with a valuation of 20 billion yuan. In September, Zhipu AI completed a 1 billion yuan funding round, with investors including Zhongguancun Science City...

Among the "AI Six Strong," ZeroOne, Bichuan AI, Zhipu AI, MoonShadow AI, and Minimax have all secured financing exceeding 100 million yuan this year. Jieyue Xingchen is also rumored to be in the midst of a new financing round with a valuation of $2 billion.

MoonShadow AI and Zhipu AI are particularly favored by capital.

Retracing MoonShadow AI's funding journey, renowned investment institutions and internet giants are prominently featured.

In June 2023, MoonShadow AI secured over $200 million in angel funding from investors including ZhenFund and Sequoia China, with a valuation of $300 million. That July, it received an A-round funding from investors such as Meituan DragonBall and BlueRun Ventures. In February 2024, it secured over $1 billion in A+ funding from investors including Sequoia China, Xiaohongshu, Alibaba, with existing shareholders participating.

Notably, this February funding round was the largest single-round financing received by a Chinese large model startup since the emergence of ChatGPT, propelling MoonShadow AI's valuation to $2.5 billion.

Capital's preference has driven up MoonShadow AI's valuation. After the August investment, its post-investment valuation reached $3.3 billion (21 billion yuan), leading the "AI Six Strong."

Zhipu AI and Bichuan AI also belong to the "20 Billion Club."

In early September, Zhipu AI completed a new round of financing worth billions of yuan, with a pre-investment valuation of 20 billion yuan, led by Zhongguancun Science City. According to Qichacha, Zhipu AI has undergone 11 funding rounds, with investors including Beijing Artificial Intelligence Industry Investment Fund, Social Security Fund Zhongguancun Independent Innovation Fund, Lightspeed China Partners, Meituan, Ant Group, Alibaba, Tencent, Xiaomi, Kingsoft, Shunwei Capital, Sequoia China, and Hillhouse Capital.

In July, Bichuan AI completed an A2-round financing of 5 billion yuan, with a post-investment valuation of 20 billion yuan. Prior to this, it announced an A1-round financing in October 2023, disclosing investors including Alibaba, Tencent, Xiaomi, and other tech giants and top investment institutions.

Within two months, three unicorns with valuations exceeding 20 billion yuan emerged in China's large model industry. Behind capital's preference for star enterprises lies some phenomena worth pondering.

Most notably, while the large model hype has surged for nearly two years, and capital chases leading unicorns, the industry as a whole remains cautious.

On one hand, large models are "difficult." From developing large models to implementing them in applications, success depends not only on financial investment and human resources but also on robust technology. On the other hand, large models are "expensive," with training costs reaching tens of millions of yuan and unclear commercialization paths, posing challenges that capital hesitates to take on.

Crucially, "20 billion" is often seen as a watershed for startups. To enter the 20 billion club, capital demands higher returns, necessitating enterprises to find self-sustainability.

In other words, when a startup's valuation reaches 20 billion, capital gives you the spotlight, and you must deliver results—make money and move fast.

2. Balancing Spending and Earning: "Rising Stars" Must Learn to Monetize

"What we're doing is challenging and requires significant financial and resource support," said Zhipu AI CEO Zhang Peng in an interview. Given the current economic climate and heavy AI investments, there's a gap between expectations and outcomes, leading to significant pressure and anxiety.

Indeed, under capital's urgent demand for returns, financing is just the first step for unicorns to secure a seat at the large model table. Learning to make money is a must.

Moving away from last year's "Hundred Models War," the large model industry is embracing application and commercialization. Similar to last year's debate on technological routes, this year's commercialization paths remain a heated topic.

At this year's Zhipu Conference, ZeroOne founder Kai-Fu Lee stated, "ZeroOne is determined to focus on To C business, not loss-making To B business."

Academician Zhang Yaqin of the Chinese Academy of Engineering, on the other hand, believes that in the embodied intelligence stage, To B applications may outpace To C, arguing that "at this stage, large models truly make money in B-end infrastructure, including chips, hardware, and servers."

Essentially, it's a debate between B-end and C-end commercialization paths for large models. One side argues that B-end applications are clearer, cover a wider range of industries, and can quickly implement multiple scenarios, while C-end competition is fierce, requiring higher time costs for breakout apps. The other side contends that intensifying industry competition leads to price wars, compressing B-end profits, while C-end offers quicker returns.

Based on this, domestic large model startups initially commercialized in two camps: those focusing on C-end business like MoonShadow AI, Bichuan AI, and ZeroOne, and those pursuing both B-end and C-end, represented by Zhipu AI and Minimax.

Regarding C-end large model applications, MoonShadow AI's Kimi is the most widely known. Launched in October 2023, Kimi became a hit with its outstanding long-text capabilities. MoonShadow AI then enhanced Kimi's long-text abilities tenfold and rapidly iterated and optimized the product.

Data shows that from December 2023 to February 2024, Kimi's monthly active users were 508,300, 1,128,500, and 2,984,600, respectively. In February 2024, user numbers were nearly six times those in December 2023.

While Kimi's popularity is undeniable, the C-end market, while closer to consumers and offering quicker returns, is crowded, with many players vying for a breakthrough. The path to a true super app is long, and no one can afford to slack.

Currently, large models' C-end revenue models are relatively limited, with subscriptions being the primary source. Other monetization models are still being explored. For instance, Kimi previously introduced a "Boost Kimi" paid option ranging from 5.2 yuan to 399 yuan, similar to a "tipping" model, to explore new commercialization avenues.

On the other hand, vendors like Zhipu AI, with faster B-end commercialization progress, focus on the large model ecosystem.

Since its inception, Zhipu AI has set its sights on catching up with OpenAI. To date, it has developed complete model products benchmarked against OpenAI's, including AI efficiency assistant Zhipu Qingyan, high-efficiency code model CodeGeeX, multi-modal understanding model CogVLM, and text-to-image model CogView.

Zhang Peng, CEO of Zhipu AI, has repeatedly emphasized that compared to the C-end market, B-end users are more willing to pay. Along this path, Zhipu AI has made many layouts around the B-end market.

For instance, it proposes the concept of "Model-as-a-Service," encapsulating large models into an open platform, providing APIs for developers and enterprises to invoke and pay based on usage. Responding to medium-to-large enterprises' data security needs, Zhipu AI offers cloud-based private deployment solutions, helping users establish dedicated model zones in the cloud.

Regardless of the path chosen, startups aim to make money but face similar challenges. C-end commercialization struggles with low user retention and high customer acquisition costs, while B-end commercialization grapples with industry price wars, putting pressure on startups.

Some vendors opt for a two-pronged B-end and C-end strategy.

In August 2024, MoonShadow AI launched Kimi's enterprise-grade API, continuing its push into the B-end market. Compared to general models catering to C-end needs, enterprise-grade model inference APIs offer higher data security and concurrency rates, supporting complex workflows and large-scale data processing within enterprises.

Meanwhile, Zhipu AI has also begun exploring C-end business development and implementation.

In July 2024, Zhipu AI launched its video generation model Qingying, which can generate a 6-second video in just 30 seconds. In August, the Zhipu Qingyan app introduced video call functionality.

Another unicorn, Minimax, also pursues a dual C-end and B-end strategy. For C-end, it offers AI chat app Glow for role-playing, immersive AI content community Xingye, and paper-writing support Hailuo AI. For B-end, it has released MoE large language models abab 6 and abab 6.5, with plans to open APIs.

Since its surge in popularity, the large model industry has passed several milestones, most notably the evolution from parameters to applications. The standard for fast-paced large models is shifting toward usability and practicality. The industry consensus is that no matter how capable general large model companies are, they ultimately rely on commercialization for self-sustainability.

Currently, almost all startup large model vendors' revenue scales are far from supporting their valuations, and applications are Trapped in homogeneous competition . Learning to make money while spending is the core of their past and future commercialization efforts, as time waits for no one.

3. Competing with Giants: Pressure on Large Model Rising Stars

In the wave of large models, internet giants and startups stand on the same starting line. If last year's technological aspect was the qualifying round, this year's application aspect has reached the finals.

It's hard to deny that startups relying solely on capital infusions face pressure from resource-rich, ecosystem-complete giants.

For startups, the primary challenges are high large model training costs and rising customer acquisition costs.

Even OpenAI, according to unpublished internal financial data cited by foreign media, faces up to $5 billion in losses this year, with estimated annual revenues between $3.5 billion and $4.5 billion but operating costs of up to $8.5 billion, including $4 billion in inference costs.

As competition intensifies and large model application user retention remains low, C-end apps are going all out to acquire new users.

Taking Kimi as an example, according to Intelligent Emergence, MoonShadow AI pays at least 30 yuan for each registered user acquired through Bilibili. Similarweb data shows that after Bilibili promotion in March 2024, Kimi's visits surged by 402.9%, outpacing Zhipu Qingyan and Minimax's Hailuo AI by an order of magnitude.

From online platforms like Bilibili, Xiaohongshu, and Douyin to offline subway stations and office buildings, the large model industry has ignited an advertising and marketing war. The results are notable, with Similarweb statistics showing a 963% surge in total visits to products from the AI Five Dragons (Zhipu AI, Minimax, Bichuan AI, ZeroOne, MoonShadow AI) within six months.

However, the cost of investing in user acquisition, which can easily exceed hundreds of millions, does not necessarily translate into long-term user engagement or spending willingness, and retention remains an unknown for large model vendors.

This year, the issue of profitability has become even more challenging. In May, vendors such as ByteDance, Alibaba, Baidu, and Tencent reduced the prices of their primary model APIs by over 90%, officially kicking off a price war in the large model market.

Some have chosen to follow suit. Zhipu AI lowered its prices twice in a month; MiniMax quietly launched a campaign offering 100 million tokens upon registration and free TPM expansion; Kimi Open Platform reduced the storage fees for context caching by 50%.

Others have chosen to hold their ground. Wang Xiaochuan, founder of Baichuan Intelligence, publicly stated that he would not follow the price cuts; Li Kaifu, CEO of 01AI, said that the crazy price cuts in the domestic large model market were a lose-lose situation for all parties involved.

Regardless of whether they choose to lower prices or not, startups must face the question: where is the next round of funding coming from?

Behind this lies the collective anxiety of startups. Without users, startups lose the data necessary for training their models, and ultimately, the enthusiasm of investors, pushing them to the brink of collapse.

Zhu Xiaohu, Managing Director of GSR Ventures, has pointed out that investing in domestic large model companies may not be profitable. He believes that an even more "awkward" situation for large model companies is that even if they are willing to invest tens of millions of dollars, they may end up wasting their money if others open-source their models.

In June of this year, a Goldman Sachs article titled "Too Much Investment, Too Little Return" bluntly stated that large companies plan to invest $1 trillion in AI-related initiatives, such as data centers, chips, and power grids, over the next few years. However, so far, these investments have only marginally improved developer productivity and have yet to yield other tangible results.

Judging from the current market response, large companies have already invested in most potential players, making it difficult for startups to secure funding from them. Coupled with limited investment market enthusiasm, startups with insufficient self-sustaining capabilities face challenges in their survival.

Drawing from the development path of foreign unicorns, it is more likely that they will be acquired by large companies. However, whether they can sell at a good price and find a good buyer remains uncertain.

Currently, Character.AI in the overseas market has sold to Google at a 50% discounted valuation, while Inflection and Adept have been acquired by Microsoft and Amazon, respectively. Meanwhile, Reka AI is still searching for a buyer, Runway has been embroiled in a "database deletion and escape" controversy, and Stability AI has reported a capital chain break after a major management shakeup.

Starting from the same starting line does not necessarily mean reaching the finish line together. Compared to large companies, even if their large model businesses incur losses, they can compensate through other businesses within their ecosystems. Startups, on the other hand, must secure funding while generating revenue and leveraging product strength to acquire customers in order to compete with large companies.

While there is unlikely to be a single dominant player in the short term, a fierce elimination round is inevitable, and every player must run their hardest this year.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.