The Chinese tech landscape is witnessing a seismic shift as major players like Baidu and Tencent recalibrate their approaches to artificial intelligence, particularly in the realm of large-scale modelsThe launch of DeepSeek has become a pivotal moment, reshaping industry norms and the competitive landscape.

On February 14, Baidu made headlines by announcing its plans to open-source its Wenxin model by June 30, a move that was quickly echoed by Tencent, with users discovering the ability to utilize DeepSeek within its WeChat search function just a day laterBoth companies, powerhouses in their own right with extensive user bases, are engaging in a scramble to adapt to an evolving market where traditional barriers to entry are being dismantled by innovations like DeepSeekThis is demonstrative of a critical vulnerability within China's tech ecosystem; despite the resources available to these giants, they find themselves scrambling to build sustainable competitive advantages around large model technologies.

Startups, too, are feeling the tremors caused by the introduction of DeepSeekPrior to the launch of DeepSeek-R1, several among China's "Big Six" startups in the large model space had already pivoted away from developing foundational modelsWith the financial landscape rapidly changing, many startups are now questioning the viability of their business models: Are their costs justified? Is their technology sufficiently advanced? Can they establish a sustainable commercial framework?

As innovation proliferates, it brings with it the harsh reality of survival of the fittest and the burden of sunk costsThe arrival of DeepSeek has led to significant valuation adjustments and heightened pressures for funding among startupsOne partner from an investment firm remarked, “Even if I can’t invest in DeepSeek, I’m not interested in you anymoreI believe DeepSeek has had a significant negative impact on their fundraising prospects.”

Then came a major pivot

Advertisements

The day before Baidu's announcement regarding the open-source model, it disclosed that its Wenxin Yiyan would be available for free starting April 1. Previously, core functionalities were offered at no cost, while customizable model training and API access came with a price tagThis transition is a marked change for a company that had staunchly advocated closed-source models, a practice similarly upheld by American counterparts like OpenAI and Google.

Closed-source strategies can bolster revenue models by keeping proprietary technology under wrapsThis fosters a level of exclusivity that can drive monetization at a faster paceHowever, the staggering capital investments necessary for large model development—often reaching hundreds of millions of dollars—have led many firms to reconsider their paths as new, open-source alternatives like DeepSeek become available.

DeepSeek disrupts the traditional closed-source narrativeIts affordability, impressive performance, and open-source nature have rendered it an attractive option for those seeking to experiment and innovate with lower operational costsMa Yue, chairman of Open Source China, articulated the strength of DeepSeek by noting, “The brilliance of DeepSeek lies in its transparency; the parameters, weights, and training processes are all shared openlyThis allows developers to leverage its methodologies to create their own models at minimal costs, which is remarkable.”

DeepSeek's innovation lies not just in its technology but in its accessibilityIt allows smaller entities to achieve sophisticated models without the necessity for massive investments that were previously deemed essentialAs one Tencent Cloud engineer noted, “Unless a closed-source model is radically superior, there’s little rationale for starting from scratch when equally capable open-source models are available.”

Companies around the globe, both closed-source and otherwise, are being compelled to reevaluate their strategies in light of these developments

Advertisements

OpenAI's CEO Sam Altman has acknowledged that choosing a closed-source approach may have placed the company on the wrong side of history, admitting that he’s yet to formulate a coherent open-source strategy for the organization.

Baidu's abrupt shift toward open-source is indicative of a broader trend in the Chinese large model industry, where immense capital expenditures and commercialization challenges are prompting many organizations to pivot.

Several years back, the so-called "Big Six" startups, such as Baichuan and Lingyi Wanhua, initially gained traction within the large model space but have now shifted their focus away from the costly challenge of developing general-purpose modelsFor instance, on January 1, 2025, Lingyi Wanhua announced its collaboration with Alibaba to establish a joint laboratory, integrating some of its team with Alibaba's workforce, transitioning its focus towards smaller models and commercial applications.

In this competitive landscape, Ma Yue emphasized the potential divisions of strategy, noting, “There are various paths; some companies in the model space are gravitating toward capital-heavy approaches while others stay committed to specific industry applications.” He referred to open-source as a “game for the brave,” highlighting that it stems from a confidence in sharing knowledge without fear of losing potential advantages.

Yet, the ascendancy of DeepSeek raises questions about whom it displacesNotably, InnoAngel Ventures, which has consciously avoided investment into any of the large model startups, views its decision as fortuitous in retrospectThe Big Six—comprising ventures such as Apollo, Minimax, and Zhigong—had at one point led the charge with technological and funding achievementsThe venture firms' typical investment ranges hover around RMB 10 million, but the soaring valuations of these startups—often starting at upwards of $300 million—presented a daunting barrier.

With giants like ChatGPT revealing transformative opportunities, domestic tech titans swiftly recognized the potential

Advertisements

Startups have found themselves competing in a high-stakes race from the very onset. “Competing directly with industry giants from the get-go can be a trap for investors,” warned Wang Sheng of InnoAngel. “Currently, there’s scant evidence of large model companies generating substantial profitsThere are avenues for revenue—such as government and enterprise projects—but these figures fall significantly short of the inflated expectations surrounding AI.”

The commercialization of AI hinges on either retrofitting existing business models or unearthing new applicationsUnlike startups, established tech behemoths like Alibaba, Baidu, and ByteDance possess operational advantages, ultimately outshining new entrants in terms of embedding AI across their existing ventures. “It’s unrealistic to expect that a startup could topple an industry giant; these corporations wield significant capital, customer bases, revenues, industry scenarios, and data,” Wang Sheng asserted. “Incremental AI innovations within established frameworks may yield opportunities, yet the likelihood of startups capitalizing on these is slim.”

New use cases remain largely uncharted territoryConsumers predominantly rely on mature hardware—smartphones and computers—for their shopping, work, and entertainment needs, which presently fulfills market demands. “Only emerging devices can foster new contextsFor example, the introduction of smart glasses could unlock new possibilities,” Wang Sheng contended, noting the current immaturity of smart glasses technology.

The outreach of DeepSeek only exacerbates the commercialization challenges faced by large model startupsNot only has DeepSeek surged past other competitors in penetrating consumer segments, with daily active users now surpassing ByteDance’s Doubao and Kimi from the Dark Side of the Moon, but it also seems poised to swiftly connect with enterprise developers.

Cloud computing platforms in China—such as Alibaba Cloud, Tencent Cloud, and Baidu Intelligent Cloud—have rapidly deployed DeepSeek, capitalizing on its open-source accessibility that eliminates the need for extensive collaboration with its creators

The allure of inexpensive, open-source models has prompted developers and consumers alike to pivot toward DeepSeek.

In a significant development, Tencent has integrated the DeepSeek-R1 model into its WeChat platform, which boasts a staggering user count of over one billionThe integration allows users to utilize DeepSeek through WeChat’s search functionality, giving them greater choices compared to previous offeringsAs Wang Sheng remarked, “Switching from one large model API to another can be a seamless process for most applications.”

Yet, these developments underscore just how difficult it has become for model companies to secure funding in a landscape now heavily influenced by DeepSeek’s emergenceSeveral investment entities have undertaken intensive discussions on the ramifications of DeepSeek for both domestic AI industry players and the broader investment environment.

Historically, launching a new model required capital investments in excess of $100 million, making it clear that success in the domain demands a significant financial commitmentInfluential players like OpenAI and Anthropic championed the notion that the pathway to successful models necessitates substantial spending, adhering to a funding paradigm borne out of exponential growth.

However, DeepSeek is shaking the foundations of this funding dogma with its presence. “DeepSeek’s innovative capabilities are unmatched among domestic startups, and it competes favorably with giants like Alibaba and ByteDance,” remarked Wang ShengThis shake-up is less about overwhelming technological superiority; instead, it compels stakeholders to re-examine the cost-to-benefit ratio of investments.

This recalibration produces palpable pressures on valuation and funding for domestic large model firms. “Can your valuation withstand scrutiny? Previous valuations relied heavily on assumptions of massive financial inputs in order to achieve success

Now’s the time to validate the necessity of billions—would $1 billion suffice? Perhaps even a few hundred million could yield satisfactory results,” Wang Sheng recommended.

For the moment, DeepSeek appears to enjoy a unique position as a startup free from funding pressuresDespite multiple investment firms vying to invest in DeepSeek, they have encountered obstacles in doing soOther startups must now navigate the dual challenge of re-evaluating their technological approaches and commercial strategies while simultaneously working to maintain funding momentum.

Yet, the mindset of investors is clearly shifting. “Even if I can’t put money into DeepSeek, I have no desire to invest in you,” predicted Wang ShengThe impact of DeepSeek on fundraising prospects cannot be overstated, presenting formidable challenges in garnering fundsHe noted that while immediate valuation corrections may feel irrational to stakeholders, they also hinder a graceful exit for investors. “They find themselves caught in a delicate game between startups and market dynamics, which complicates the path forward considerably.”

Enormous financial waste now looms as a palpable riskWhen an open-source model that is both cost-effective and high-performing emerges, companies unable to deliver superior offerings swiftly see their investments dissipateMany firms that previously invested heavily in development, pre-training, and computational infrastructure face the grim prospect of these ventures amounting to naught.

Ma Yue observed, “In China, the sunk costs can be astronomicalA company might burn through billions, only for entities like DeepSeek to emerge as the ultimate beneficiariesFurthermore, the fundamental issue lies in the vast disparity in AI investments between China and the U.S.”

The cycle of investment and innovation is unlikely to slow down, howeverBy 2025, investors may find themselves gravitating towards innovative applications built on top of existing large models

Advertisements

Advertisements