In the ever-evolving landscape of artificial intelligence, the pace of advancement can feel akin to the metaphorical saying, "One day in AI is like a year in the human world." The trajectory of AI models in 2024 showcases rapid developments that are reminiscent of a technological renaissanceOver the past year, several predictions regarding AI have come to fruitionAmidst a surging demand for intelligent computing, tech giant NVIDIA has comfortably breached the $3 trillion market cap milestone, a clear illustration of the burgeoning value of AI-driven technologiesConcurrently, China's open-source community has gained ground, with models like DeepSeek pushing the boundaries against formidable opponents like LlamaOn the domestic front, the competition in AI video generation capabilities has intensified, with fierce rivals like Sora making significant stridesAdditionally, major players in both the smartphone and PC industries are entering a frenzied cycle of hardware investment to keep up with the relentless pace of AI innovation.

Among the various competitors, ByteDance—a company previously characterized by its low profile in the AI race—has silently emerged as a formidable player with its model, Doubao

Doubao has quickly ascended to remarkable heights, securing the title of the second-largest AI model globally while reigning supreme in ChinaThis unexpected triumph has left many competitors in awe, particularly considering ByteDance's seemingly late start in the AI arena compared to its rivalsDespite a slow initial development phase, Doubao has surged ahead, overtaking numerous established players in a surprisingly short period, cementing its status as a powerful competitor.

So, how has Doubao, as a relative latecomer, managed to carve out its niche and come out on top? In November of this year, Doubao's app recorded an astonishing 59.98 million monthly active users (MAU), making it second only to OpenAI's ChatGPT and dominating the domestic marketThe app's international counterpart, Cici, also boasts an impressive MAU of 12.67 million, placing it 22nd globallyRemarkably, Doubao’s user base was just 1.73 million at the beginning of the year, indicating a meteoric rise

• Fed's Warning Drags Down Regional Bank Profits
• 5% Dip: Is the Safe-Haven Rally Over for Gold and Oil?
• Dollar-Yen Falls Below 155
• Cutting Prices Won't Boost Automaker Competitiveness
• Mortgage Rates Move with Supply and Demand

The question now stands: what propelled Doubao's rapid ascent?

The first factor contributing to Doubao's success is ByteDance's internal consensus to provide extensive support across resources, talent, and mechanisms for the development of the Doubao modelIn response to this AI technological revolution, ByteDance swiftly recalibrated its strategic framework, implementing a series of measures that included large-scale recruitment of AI personnel, the establishment of a dedicated AI department known as Flow, the integration of internal resources, and the adoption of competitive strategies to accelerate Doubao's development.

Such concerted efforts and investments culminated in significant achievements by March of this year, when Doubao's downloads and monthly active users both reached unprecedented heightsThis upward trend continued into November, despite ChatGPT still leading with over 300 million MAU across app and web users

However, Doubao's growth rate far surpasses that of its competitorsDomestic applications like Kimi, Wenxiaoyan, and Tongyi Qianwen have been left trailing behind, with Doubao's active user base even exceeding the combined total of these three rivalsByteDance has notably displayed an aptitude for timing, adeptly leveraging open-source resources and mobilizing favorable conditions to swiftly translate new technologies into tangible outcomes.

Secondly, Doubao's strategy has been characterized by aggressively outpacing competitors through competitive pricing and user engagementWith the ongoing fray of domestic AI models, a fierce battle concerning technological iteration and commercialization efficiency has eruptedTo compensate for its initial delay, ByteDance has spared no expense in a comprehensive approach to competitionThis has manifested through substantial investments aimed at boosting user engagement on the consumer side, while also reducing prices on the business end and competing on computing power.

For instance, ByteDance reportedly funneled no less than ¥1 billion (approximately $150 million) into consumer-facing marketing to accelerate the growth of Doubao's user base

This intense marketing push has ensured Doubao's visibility on social media platforms, search engines, and short video apps, thereby establishing a robust brand recognition in consumers’ mindsIn the B2B realm, Doubao has introduced price cuts that have taken competitors by surpriseDuring the Vulkan Engine FORCE Live Power Conference this year, Doubao's Pro 32k model was priced at ¥0.008 (roughly $0.001) per 1,000 tokens, while the Pro 128k model was set at ¥0.05 for the same volumeThis aggressive pricing strategy ushered in a new era for domestic models, leaving rivals scrambling to reactively adjust their own pricing.

Moreover, the competition for dominance among major models is fundamentally a contest of computational powerByteDance has reportedly committed a staggering ¥80 billion (approx$12 billion) to its AI capital expenditures—figures so substantial they rival those of the leading tech giants (total combined expenditures approximating ¥100 billion). The latest disclosures from ByteDance indicate that capital expenditures will expand to ¥160 billion by 2025, earmarking ¥90 billion for AI computing clusters and ¥70 billion for data center and networking infrastructure

Such a comprehensive and saturated offensive approach is the primary reason for ByteDance's rapid rise to prominence.

As domestic mega-models continue to evolve, a host of companies are springing up within this booming sectorThe year 2024 has not only witnessed the emergence of the "Six Small Tigers," a group of domestic models that have collectively generated ¥20 billion, but significant investments are also being funneled into AI hardware companies that integrate AI models into their offeringsAccording to public data, as of November 2024, daily API call volumes for Baidu's Wenxin model surpassed 1.5 billion—an increase of approximately 30 times compared to just over a year priorAs of mid-December 2024, the daily token usage of Doubao's general model exceeded 4 trillion, marking a 33-fold increase since its debut seven months earlier.

Nevertheless, changes on the horizon pose challenges for many previously successful competitors, the "Six Small Tigers" now facing a reality check with funds drying up and commercialization efforts encountering steep obstacles

The competitive landscape has shifted dramatically; newly advanced companies are rising, while established giants ramp up their offensive strategiesThis ebb and flow of market dynamics recalls the age-old adage: the rich get richer, leading to a resurgence of winners-take-all scenarios amidst AI model competition.

Mainstream tech giants now possess significant advantages in terms of users, applications, and computational resourcesFor instance, Baidu leverages its expansive APP ecosystem and vast search engine capabilities alongside its cloud services; Alibaba benefits from a plethora of applications including Alipay, Taobao, Tmall, and Cainiao, all supported by its cloud infrastructureSimilar advantages are equally enjoyed by Tencent and ByteDanceMajor corporations with luxurious resources can seamlessly navigate the realm of mega-model development.

In contrast, many startups find themselves reliant on third-party platforms for critical data resources

This dependence translates to high marketing expenditures and exposes them to competitive vulnerabilitiesData from mobile marketing platform AppGrowing indicates that since the beginning of 2024, Kimi, Doubao, and several other domestic models collectively engaged in over 6.25 million advertising placements at a cost exceeding ¥1.5 billionNevertheless, established companies benefit from direct access to traffic channels, reducing the need for such levels of investment and allowing them to crush smaller competitors simply by leveraging their inherent traffic advantages.

An illustrative case is Kimi—once a rising star that captured massive attention on platforms like Douyin and Bilibili but found itself restricted from advertising on Douyin in the latter half of the year, prompting a shift to alternatives like KuaishouData from AI xinxibu shows that Kimi's ad placements had exceeded 2,500 in the past three months, overwhelming the competition as over 2,100 of these were made on Kuaishou alone, comprising 84% of its total placements

In B2B markets, the advantage for larger firms is even more pronounced; they can offer tokens at significantly lower costs, enabling them to entice business users to experiment at minimal expense—a luxury not afforded to fledgling startups.

Additionally, technologically advanced firms and major corporations are crushing the survival opportunities for startups through the tightening of business modelsMany AI model providers rely heavily on API sales as a key revenue streamProjections for 2024 suggest that OpenAI will earn around $500 million from API sales, contributing 15% to its total revenueLikewise, Anthropic is anticipated to derive a staggering 60% to 75% of its total revenue from third-party API usage.

However, the current dynamics within the domestic AI model landscape render API provision as a battleground for market share rather than a source of generous profit margins

In May 2024, DeepSeek announced its DeepSeek-V2, with exceptionally low pricing of just ¥1 per million input tokens and ¥2 for output tokensDoubao promptly followed, bringing their rates down to unprecedented levelsIn response to the resulting price wars, companies like Alibaba's Tongyi Qianwen drastically slashed prices by 97%, and Baidu opted to offer two of its models for free.

This drastic price competition has left many startups corneredFounders of their companies, such as Wang Xiaochuan of Baichuan Intelligent, commented that they would not participate in this price war, while Li Kaifu of Zero One Technology warned, "If technology is lacking and businesses engage solely in price cutting, there will be no winners." In private deployments for major enterprises, startups are at a distinct disadvantage, with major successes predominantly claimed by established players such as Baidu, Tencent, Huawei, and major telecom companies

Efforts from startups like Zhipu AI have only yielded a few minor wins.

As the landscape continues to evolve, it becomes evident that both technical prowess and available resources will dictate the fate of competitors, with stronger entities squeezing out smaller players in terms of viabilityIf this trajectory continues, we may witness a recurrence of the winners-take-all mentality within this domain.

Shifting gears further into 2024, the reshaping of AI models enters an accelerated phaseMuch like the evolution of other emerging technologies, early stages of model development were characterized by an insatiable thirst for innovation, even at the cost of resource depletionHowever, as various models proliferate across the market, the entire ecosystem is on the verge of a transformation, prioritizing commercialization and the advancement of cutting-edge technology, thus propelling market clearing.

Firstly, the intensifying spending on technological research and development coupled with difficulties in financing compel various manufacturers to hasten their pace towards commercializing mega-models

Startups that previously enjoyed seamless financing options are now experiencing challenges, propelling them to adapt rapidly to market conditionsIn contrast, although the substantial capital investments required by larger corporations don’t threaten their foundational operations, they do represent a meaningful financial outlay.

Moreover, technological advancements in model efficiency are also evidentThe rapid decline in inference costs and the increasing popularity of lightweight models and Mixture-of-Experts (MoE) models have facilitated lower-cost and higher-performance models emergingThis is observable on the ground, where the days of multi-million dollar projects are dwindling; now, many applications are becoming viable at two to three million yuan, and some targeted models can be deployed with initial investments as low as a few hundred thousand yuan, making them attractive to potential clients.

There are predictions that by 2025, as inference costs continue to diminish, the deployment of mega-models will pick up even more momentum

Hard data supports this notion; records show that over the first eleven months of this year, the number of mega-model projects has soared to 728—3.6 times compared to all of last yearCommercial ventures have also transitioned from a broad approach directed towards the entire industry to a focused exploration within specific sectors; the service delivery method has shifted from simple transactional buying of units to a focus on application outcomes and a more integrated service model.

Secondly, the evolution of mega-models necessitates a focus on delivering substantial value, as models without significant functional improvements risk becoming obsolete and being cast aside by usersThroughout 2024, the fierce competition has led to a settling of foundational mega-models both domestically and abroadFor instance, the American scene is now dominated solely by OpenAI, Anthropic, Meta, Google, and xAI, while China’s landscape has consolidated around a handful of players such as Baidu, Alibaba, Zhipu, ByteDance, iFlyTek, and Tencent.

Essentially, after a year of fierce contests, the arbitrage opportunities surrounding foundational models have diminished, and to stand out moving forward, creating impactful AI applications is essential