On November 29, during the Agent OpenDay event, a curious journalist posed a question to CEO Zhang Peng regarding the company's progress in the B2B sectorZhang's response was somewhat terse, simply stating, "It's going okay." This brief remark left many pondering the challenges that lie ahead for the AI enterprise.

This year has seen the rise of emerging AI companies, most prominently participating in the cutthroat competition within the burgeoning field of large modelsFacing colossal competitors like Baidu, Alibaba, Tencent, and ByteDance, it appears that the fortunes of these up-and-coming AI firms mirror one another: getting shortlisted for major contracts has become a consolation prize more than a solid victoryIn most instances, these companies walk away with the acknowledgment of participation rather than substantial success.

The battle against these tech giants has proved to be particularly challenging

As of December 3, a search using the term "large model" on China's public procurement platform unearthed around 200 contracts awardedFurthermore, a preliminary tally by industry analysts indicates that the four leading cloud providers won a staggering 98 contracts, representing nearly half the total market shareCollectively, these contracts amount to 1.112 billion yuan ($150 million), a hefty chunk of which has been claimed by state-owned enterprises and government entities.

In this whirlwind of activity, AI companies find themselves ensnared in a self-perpetuating cycle, needing to constantly drum up interest through promotional events and product launches to entice B2B clientsThis noise, however, is overshadowed by the inherent advantages of larger cloud corporationsTheir pricing for computing power orders tends to dominate, often undercutting competitors by significant marginsMeanwhile, the pure functionality of large models doesn't justify high price tags when bundled with cloud architecture and database solutions, which can be repackaged to offset expenses and enhance negotiation leverage

• Boosting the Economy with Fiscal Policy
• Gold and Silver Prices Fall Today
• Gold Prices Hit Two-Week High
• Australian Iron Ore Prices Fall
• Gold Prices Hit One-Month High

Customers frequently exhibit a path dependency, gravitating towards familiar cloud providers with whom they've previously collaborated.

The outlook is becoming increasingly worrisomeIn the B2B market, cloud service providers have initiated aggressive price wars, offering increased services without raising costs, and even underbidding competitors by less than half their prices in some situationsMeanwhile, the early advantages of many AI startups are waning; their offerings are becoming indistinguishable, and they often fall short compared to established players in terms of qualitative services.

This landscape has essentially set the stage for 2024: the big clouds will feast while the AI startups are left to sip the soup.

Turning to the figures related to AI revenue in financial reports, it becomes evident that starting from last year, cloud service giants have elevated AI and large model projects to strategic significance

Individual strategies vary, with Alibaba Cloud, for instance, embracing a Model-as-a-Service (MaaS) approach aimed at empowering cloud revenue through its AI products and large models.

In the current fiscal structure of Alibaba Cloud, revenue is derived from five main areas: the utilization of its Tongyi Qianwen model by internal businesses within the Alibaba ecosystem, client purchases and consumption of cloud services from affiliated products like DingTalk, income generated from AI companies supported by Alibaba, tokens revenue from developers utilizing the large models, and income from B2B market ordersAlibaba Cloud has emerged as the industry leader, covering a diverse range of cloud service offerings, and has seemed to weather market trials across both open-source and proprietary paradigms.

Leveraging these multiple revenue streams, Alibaba Cloud has reported a surge in both revenue and profit, with figures jumping from 25.595 billion yuan to 29.61 billion yuan from Q1 to Q3 of 2024, and adjusted profits rising from 1.432 billion yuan to 2.661 billion yuan—an impressive year-on-year growth of 89%. Unfortunately, despite this success, Alibaba Cloud has yet to achieve double-digit growth annually.

Conversely, Baidu has taken a more aggressive stance on AI and large models, positioning itself as a leader from model release to AI transformation

Although its IaaS and PaaS foundations are not as robust as Alibaba's, Baidu has shifted focus away from infrastructure towards model and application developmentBaidu Cloud’s revenue sources primarily include token income generated from developers accessing the Wenxin model, fees from authorized API usage, and revenues from B2B orders.

The absence of economies of scale in basic infrastructure has compelled Baidu to pursue B2B opportunities vigorouslyMultiple startup founders have reported that Baidu’s marketing teams proactively engage with them to facilitate collaborations with other startups—an effort aimed at rapidly securing orders and boosting Wenxin token utilization.

Baidu Cloud has seen the most pronounced growth, with their 2024 Q3 financial report indicating that AI revenues have grown to account for over 11% of total income, largely propelled by high demand for model training and inference in sectors such as internet services, education, and finance

Notably, incremental income from mid-tier enterprises surged by 170% quarter-over-quarter.

However, as similar products continue to appear in the market, Baidu's uniqueness is increasingly under threatWhile revenue continues to climb, the pace of growth is noticeably slowingThe Q3 sequential growth rate plummeted from 14% to 11%, with generative AI cloud revenue dipping sharply from 95% to just 17%.

In stark contrast, Tencent has adopted a more conservative approach, preferring to enter the market when industry conditions become clearerThis applies to developments in large models, AI assistants, generative images, and AI video technologiesUnlike Alibaba and Baidu's race for new customers, Tencent's strategy revolves around leveraging AI to meet existing business needs, making integration into their ecosystem a priority.

AI revenue for Tencent is dispersed across various business segments, similar to scattered pebbles resulting in progressive ripples that are hard to quantify in the short term

According to Tencent’s Q3 report, the marketing services division’s growth has been buoyed by revenue from Video Accounts, mini-programs, WeChat advertising, and AI technologyCurrently, AI revenue constitutes around 10% of Tencent Cloud’s total income, with predictions for notable free cash flow generated by AI business in the upcoming year.

The landscape for the large model market in China is still developing but filled with uncertaintiesCustomers remain hesitant, and the infrastructure to support large models and AI functionalities is incomplete, with developers frequently jumping from one provider to another in search of better opportunitiesPresently, in the broader Chinese economic setting, sustained and steady demand for large models is primarily emanating from government and enterprise sectorsWhile vendors claim they are targeting the B2B market, the reality is that they predominantly serve governmental entities.

For cloud service providers, the government market offers access to pre-existing channels, experience, and clientele

Essentially, they are repeating previous processes, with the focal points shifting from SaaS and cloud to large models and AI AgentsConsequently, landing contracts has become the hallmark of cloud companies frantically racking up points this year.

Using openly accessible data from various platforms, researchers have tracked contract awards related to large models by key players, such as Baidu Cloud, Alibaba Cloud, Tencent Cloud, and Volcano Cloud, up to December 3, 2024. It should be noted that this data focuses exclusively on orders generated by large models and generative AI.

Among these four, Baidu Cloud leads in contract acquisition with 34 awards, predominantly in the financial sector, followed by telecommunications, electric power, scientific research, environmental management, and public servicesThey secured a total monetary value of 446 million yuan, with computing power and large model application services making up the bulk of this figure at 216 million yuan and 170 million yuan, respectively

While this hunting success confirms Baidu’s mental edge in AI search, the average contract size remains modest, capping at no more than 2 million yuan.

Alibaba Cloud managed to win 18 contracts, with a major focus on finance, educational research, and governmental sectorsAlthough they trail Baidu in number, they closed in with a total of 426 million yuan due to robust revenue from intelligent computingAlongside computing power, their large model training and deployment contracts were significantly smaller at 12.96 million yuan and 7.69 million yuan.

Analyzing this data indicates that Alibaba’s strengths still lie in their traditional cloud business, which includes services like scaling, public cloud offerings, databases, middleware, and hardware purchasesNew ventures, such as digital humans, smart customer service, and AI programming, represent a smaller proportion of their portfolio, with Alibaba’s large model acquisition derived from their investments in the Zhijiang Laboratory, a research institution.

Tencent Cloud ranked third in contract wins with a total of 24 awards, amounting to approximately 180 million yuan, focusing on media and telecommunications

Insights from the contract details reveal a clear understanding among governmental and corporate entities regarding Tencent's mixed cloud offerings, combining digital humans and multi-modal processingThis probably stems from the association with Tencent’s MoE technology and an open-source strategy, bolstering its bid success in large model training at 72%.

Volcano Cloud secured 22 contracts but amassed only 61.59 million yuan in totalThey garnered orders across various specialized AI fields; due to the less complex nature and lower customization demands of intelligence agents, average contract prices tend to fluctuate.

Among five contracts related to intelligence agents, the minimum size was 480,000 yuan, while the maximum transaction was 4.2 million yuanThis pricing render is relatively low, indicating an evident market shift from the C-end to a more B-centric approach, aiming to bridge between large models and B2B functionality

Additionally, the audio-visual capabilities of the Doubao model have also attracted attention.

Overall, despite the similarities in the awarded industries and financial distributions among the cloud vendors, the competitive landscape remains formidableUnless they are awarded substantial computing contracts, it is challenging to establish any single provider as overwhelmingly dominantAcross the board, cloud providers face the common dilemma of excessive competition with insufficient opportunities, rendering the actual average value of large model and related AI contracts disappointingly low.

When computing power is excluded, the average contract value breaks down to approximately 7.42 million yuan for Baidu Cloud, around 2 million yuan for Alibaba Cloud, approximately 7.5 million yuan for Tencent Cloud, and around 2.8 million yuan for Volcano CloudThese figures blur into a complex picture grounded in the distribution of profits among contractors; following typical practices of significant enterprises, contracts are often divided into multiple components distributed among various companies.

Amidst the large number of awarded contracts, many procurement announcements list "single procurement sources," implying that success frequently rests with established cloud providers

This indicates that many AI companies are essentially sidelined from the very beginning.

This trend highlights a significant long-standing resource monopoly by cloud providers in this segmentMany state-owned enterprises prefer to stick with established vendors for their AI and large model services, driven by factors such as security, continuity, and stability, thereby minimizing migration and development costs.

As B2B or B2G enterprises, the essential needs often revolve around providing comprehensive, project-based solutionsClients may simultaneously require hardware, cloud services, extensive model training, and agent application development—essentially, a well-rounded custom solution encompassing hardware and softwareFor example, in certain educational digitization contracts, the procurement lists demand cloud expansion, customized large model invocation services, document tools, office software, and data analytics.

For new entrants attempting to penetrate such industries, their existing products likely fall short of meeting comprehensive client needs, further sidelining them from pursuing meaningful contracts

Opportunities for AI firms to compete alongside the giants often arise when tenders are sliced into various components—like generating text, images, music, or videos—allowing a fair comparison between startup capabilities and those of more established counterpartsDespite an underwhelming performance in large multiservice contracts, companies like Zhichu have still managed to snag a few projects involving multimodal applications, large model training, pre-training, and AI video applications amid fierce competition.

Inadvertently, this situation nudges many AI startups towards a "complementary" approach—most leading edges have diversified into large model training, inference, AI search, generative images, and AI videoAs they become increasingly homogenized, the discrepancies between them and larger firms have also begun to blur; the giants continue to evolve, delivering more standard and reliable services

This year, major cloud companies seized nearly half of the available contracts in the large model and AI market, while innovative firms like Zhichu only manage to sip the soup, leaving the majority to go hungry.

The lengthy delivery cycles and slow billing processes in government contracts alone are enough to drain a number of struggling AI companies on the brink of collapseFor the giants, large model contracts represent an extra feather in their cap, while for startups, they make up critical income—it is a dichotomy of necessityThe financially secure enterprises can endure, engaging in aggressive price battles at the expense of profitabilityCoupled with the consistently declining average price points for large model and AI contracts, major cloud providers linger as partial culprits; in competitive scenarios, they often underbid to secure orders.

In a recent bidding contest for a “domestically developed intelligent programming service project,” a competitor quoted 980,000 yuan, while another offered 560,000 yuan, whereas Alibaba undercut the competition, sealing a deal at 350,000 yuan