In recent years, the field of artificial intelligence (AI) has undergone remarkable transformations, with a plethora of new technologies and models emerging at an unprecedented paceThese advancements have sparked a series of revolutionary changes in various industriesOne significant player in this evolution is DeepSeek, which has initiated an efficiency revolution within the AI domainHowever, it provokes an unexpected reaction regarding the demand for GPUs (graphics processing units) in the AI sector; the appetite for GPUs appears to remain robust and potentially increasing.

An intriguing viewpoint comes from Andrei Mida, a partner at the leading Silicon Valley venture capital firm a16z and a board member of the French open-source AI company MistralDrawing from his extensive experience and keen insights in the AI landscape, Mida has unique perspectives on DeepSeek's efficiency revolutionHe firmly believes that the efficiency improvements brought by DeepSeek will not lessen the AI industry's ongoing GPU 'hunger games'; instead, they will significantly escalate computational needs driven by enhanced model efficiency.

Mida points out that when computational output improves tenfold, companies don't simply halt their chip purchases

Advertisements

On the contrary, they leverage equivalent resources to accomplish a greater array of tasksIn the fiercely competitive AI market today, businesses are racing against time to seize opportunities, expand their fields of operation, and enhance product performanceComputational power is the core driving force behind AI's development—analogous to fuel for industriesGaining more computational power equates to obtaining a competitive edgeTherefore, when output capabilities surge, organizations are quick to funnel those resources into additional projects, seeking to unlock further commercial value and innovation prospects.

Reflecting back to six months ago, Mida expressed his astonishment at the code generation model Coder V released by DeepSeekThis model showcased remarkable advantages in code generation's efficiency, accuracy, and intelligence, fostering new possibilities within the AI programming landscapeCoder V's arrival has not only transformed traditional perceptions of code generation but has also garnered the AI community's admiration for DeepSeek's technological prowess.

Using Mistral as an example, Mida elaborates on how DeepSeek's rise impacts its peers within the industryWith Mistral having raised a staggering $1 billion in funding, one might ponder whether this investment could be perceived as excessive in light of DeepSeek’s ascensionMida, however, presents a contrasting viewpointHe asserts that being a competitor allows Mistral to witness DeepSeek's efficiency improvements firsthand and subsequently internalize these insights before investing the $1 billion—an invaluable opportunityHe emphasizes that with the same computational power, tenfold output can be achieved, enabling businesses to leverage DeepSeek's enhancements to generate higher productivity and profits without incurring significant additional costs.

Mida’s understanding of the competitive logic in the open-source ecosystem is profound

Advertisements

He posits that in the realm of open source, corporations do not need to secure staggering amounts of capital—what truly matters is possessing more computational power than any rivalComputational power functions like a 'hard currency' within the open-source sphere; possessing superior computational capacity affords a company the advantage in model training, optimization, and innovation endeavors, thereby attracting more developers and users, and firmly establishing a competitive foothold in an intense market landscapeThis elucidates why OpenAI continues to partner with entities like SoftBank and Oracle for the "Stargate" super-datacenter initiativeThe purpose of the Stargate project is to create a super-datacenter endowed with robust computational power, thus providing a solid support base for OpenAI's research and business evolution, ensuring its frontrunner status in the AI arena.

For Mistral, a pivotal weapon in competing against closed-source giants like OpenAI and Anthropic lies in its open-source strategyOpen-source models possess distinct advantages that attract developers worldwide, fostering a collaborative environment where free technological labor is harnessed into a large open-source communityIn this ecosystem, developers work together, share experiences and technological breakthroughs, and continually push for model optimization and innovationIn stark contrast, closed-source firms face the burden of shouldering their research and computational expenses alone, which inevitably intensifies operational pressures and risksCurrently, Mistral boasts that it has amassed the "largest computational reserves in the open-source domain," a distinct advantage that provides robust backing in its competitive strategies.

As the head of a16z's GPU sharing program, Oxygen, Mida is acutely aware of the current shortage of computational power

Advertisements

The Oxygen project invested heavily in Nvidia H100 chips a year and a half earlier, designed to provide computational support for its portfolio companiesHowever, demand now exceeds supply, with orders having "dramatically overshot" the available capacityMida laments this situation, stating, "Training models necessitates GPUs, while operational products require sustained reasoning power—this is an endless pit." This starkly illustrates the AI industry's ongoing and massive demand for computational resourcesAs AI technologies further evolve and application scenarios expand, this need is set to grow continuously.

When queried about whether any companies had shelved their data center plans due to the rise of DeepSeek, Mida humorously remarked, "If anyone has idle GPUs, please send them to me." This seemingly jesting comment underscores the acute demand for computational resources within the AI industryIn the current market climate, computational power has become a critical asset for corporate development; those with more power gain the upper hand in the competitive landscape.

The efficiency revolution sparked by DeepSeek undoubtedly ushers in fresh opportunities and momentum for the industryHowever, it does not alleviate the AI sector's robust demand for GPUsOn the contrary, as model efficiency enhances and application scenarios proliferate, the demand for computational resources will continue to soarFor businesses, effectively managing computational resources and maximizing their utilization efficiency within limited means will be key to future advancementFurthermore, competition within the open-source ecosystem is set to intensify, compelling companies to consistently innovate and refine their strategies to face myriad challenges.

As we look ahead, how will the landscape of computational power in the AI sector evolve? What fresh changes will DeepSeek’s efficiency revolution introduce to the industry? These inquiries deserve our sustained attention and deep contemplation

Advertisements

Advertisements