The financial reporting season for major tech companies in the U.Scoincided with the striking rise of the Chinese AI model DeepSeekThis unexpected timing has ignited discussions within the industry regarding the potential impact of DeepSeek on AI investments and the landscape of cloud services.

In the wake of DeepSeek's emergence, executives from leading firms like Microsoft, Google, Meta, as well as AI chip manufacturers like AMD and Arm, found themselves fielding questions about how this new model would influence their strategies and investmentsWithout exception, these industry titans voiced their admiration for DeepSeek, revealing their intentions to leverage its breakthroughs to enhance their own technologies and modelsHowever, despite mounting skepticism surrounding the substantial AI investments of these companies, they maintained their commitment to substantial financial outlays, perceiving vast uncertainties and potential for exploration, particularly in application development.

DeepSeek's rise has significantly narrowed the competitive divide between open-source and closed-source AI models, and it stands to expedite the evolution of ecosystems surrounding open-source frameworks

Advertisements

How might DeepSeek reshape the strategic perspectives of the giants of technology?

The launch of DeepSeek has prompted both scrutiny of its technical roadmap and recognition of the expansive market potential unlocked by decreasing costs associated with AI model deploymentMark Zuckerberg, the CEO of Meta, emphasized that while open-source offerings initially lacked competitive edge, greater transparency could foster industry-wide standardization, motivating collective efforts to streamline costs and enhance capabilities.

"DeepSeek, as a new competitor from China, signals the imminent establishment of a global open-source standard—one of the pivotal topics under discussion," Zuckerberg remarkedHe noted that DeepSeek embodies a plethora of advanced ideas, which Meta aims to absorb and potentially apply within its own systems"Innovation in technology often stems from continuous improvements led by various companies, which then serve as a resource for others to learn and evolve," he added.

In light of DeepSeek's rapid success, several tech giants, including Amazon and Microsoft, have quickly announced their intention to integrate DeepSeek's model API

Advertisements

This swift embrace highlights the accelerating potential for AI applications fueled by significantly reduced inference costsMany industry leaders have articulated optimism regarding the developmental opportunities AI inference brings, necessitating strategic readiness.

Amazon's CEO, Andy Jassy, expressed his admiration for DeepSeek, particularly noting its innovations in training optimization techniques, such as reinforcement learning"For those committed to building cutting-edge models, we are all studying similar methodologies and learning from one anotherIt's evident, and will continue to be the case, that our companies will frequently surpass one anotherExpect a rich array of innovations to emerge as a result," he stated.

Similarly, Arm's CEO, Rene Haas, acknowledged that DeepSeek's various models, including V3 and R1, showcase significant creative efforts built upon existing industry-leading frameworks that make inference more efficient

Advertisements

"Frankly, I think it's fantastic," he commented, recognizing that this progress could help the industry achieve greater efficiency, reduce costs, and cater to rising computational demands.

Sundar Pichai, CEO of Google, characterized DeepSeek as an exceptional teamReflecting on the past three years, he noted an increasing proportion of spending directed towards AI inference compared to AI training"This is certainly a positive trend as inference can help businesses secure strong returns on investment, accelerating the deployment of applications," he affirmed.

Pichai also stated that the costs associated with inference will continually decline, thus opening up a plethora of viable new use cases"The scope of opportunity is vast, which is why we are continuously investing to prepare for this moment," he remarked.

The swift evolution in demand for AI inference capabilities is an undeniable reality

• The Inability of Jinshan Cloud to Stand Alone
• SK On: A Strategic Shift with Strong Prospects
• The Rise of A-Shares and the AI Investment Boom
• The UK’s Economic Dilemma: Inflation
• The Golden Era of Gold

However, DeepSeek's ability to deliver these capabilities at lower costs introduced a noticeable contrast with the staggering AI investments often reported by U.Stech giants, which can reach into the hundreds of billions.

During earnings calls, discussions around AI investment trajectories drew significant attentionCollectively, these companies reasserted their commitment to investing in AI infrastructure while highlighting the critical value of investing in inference capabilities—particularly for application explorationJassy opined that recent weeks have led to assumptions about reducing overall tech spending by minimizing the cost of any AI technology component, most notably inference.

However, he cautioned that this assumption might not hold trueDrawing from his experience in cloud computing, Jassy recounted that when Amazon Web Services launched in 2006, its S3 storage service was priced at 15 cents per gigabyte while computation cost 10 cents per hour—fairly cheap compared to today’s standards

"As technology progresses, an initial belief arose that corporate expenditure on infrastructure tech would dramatically decreaseWhile it's true that spending per unit of infrastructure significantly reduces, companies often become intrigued by new projects they had previously deemed too costly, ultimately leading to greater overall expenditures," he explainedJassy concluded that while inference costs are expected to drop, this trend will positively benefit customers and the business overall.

Meta's CFO, Susan Li, affirmed the company's continued dedication to infrastructure investments for both AI training and inference"It remains unclear what we specifically need, including the breadth of our inference applications, which is fundamentally our competitive advantageThus, we are quite excited, as there remains substantial space to advance the efficient operation of these workloads," she noted.

"I still believe that over time, substantial capital expenditures and infrastructural investments will present a strategic advantage in terms of service quality and scale," Li stated

She elaborated that for 2025, Meta's budget for AI infrastructure would predominantly focus on GPU deployment, coupled with ongoing enhancements in network capacity and optical transmission capabilities.

Li maintained, "It's premature to ascertain the long-term capital efficiencyA multitude of factors, including the pace of foundational model advancements, the direction of GenAI product use cases, and the performance improvements afforded by next-generation hardware innovations, must first be considered." This aligns with the notion that in a naturally evolving industry landscape, early movers often need to invest heavily in exploration, thus granting later entrants some cost advantages.

Nevertheless, the ascendance of DeepSeek has created a sharper contrast in this dynamic, although it cannot be assumed that the tech giants' AI investments are purely "wasted," as they may simply face accessibility and efficacy challenges.

Delving into the technological aspects, DeepSeek's release has led to a significant drop in the stock prices of numerous chip manufacturers, prompting some to argue that the need for AI chips is diminishing

However, this perspective overlooks the reality that, from another angle, DeepSeek could indeed enhance opportunities for AI chip developers.

NVIDIA previously indicated a rapidly growing demand for AI inference within their earnings discussionsThus, it would be unreasonable to conclude that the relevance of NVIDIA's GPU cards is waning based on current trends.

In recent developments, several domestic chip manufacturers—including Huawei's Ascend, Moore Threads, Muxi, Birran Technology, and Tianxu Zhixin—have pursued adaptations to DeepSeek's two models.

The implications are clear: advanced AI models are likely to enable a host of devices to gain computational capabilities, which will substantially amplify the demand for inference-level computingThis expansion is akin to how AI glasses are anticipated to succeed traditional headphones and glasses as new product categories—an evolution that will gradually unfold as the applications broaden.

It is crucial to recognize, however, that in the AI era, empowering more devices with computational capabilities imposes stringent demands on chip technology

Understanding the trajectory of technological advancements will be vital for chip vendors.

Rene Haas analyzed that given the diverse environments in which AI workloads must operate, achieving low-cost, high-efficiency inference is essential, especially for products constrained by power limitations"I believe DeepSeek is actively contributing to heightened computational needs, positioned as a valuable opportunity for Arm to enhance efficiency," he stated.

"Yet, products like the Grace Blackwell series chips (collaboratively designed by NVIDIA and Arm for high-performance data center applications) currently face challenges in applying directly to mobile devices or wearables, including headsets and automobilesThere remains substantial work to be done here, with vast space for optimization," he elaborated.

Haas expressed that the ongoing substantial investments made by tech giants in AI stem from a consensus that "we are still far from having the capability to bring transformative applications in the realm of artificial intelligence." Alongside improving edge-efficiency, harmonizing software and hardware capabilities is equally important.

In a recent technology trend report, Arm noted the increasing complexity of chips and software, emphasizing that no singular company can independently design, develop, or integrate all aspects of the required chip and software

Thus, robust collaborative efforts within ecosystems are imperativeSuch cooperation enables individual companies to provide differentiated computational components and solutions according to competitive strengthsFor instance, the automotive industry necessitates the cohesive integration of chip suppliers, primary suppliers, vehicle manufacturers, and software providers into an ecosystem that maximizes AI's potential for end users.

Satya Nadella, CEO of Microsoft, also highlighted that the company is actively establishing a versatile pool of computational resources to ensure a balanced approach between training and inference"We have invested heavily in software optimization, which not only includes leveraging DeepSeek's successes but encompasses our longstanding collaboration with OpenAI aimed at cost reduction for GPT models," he saidHe asserted the growing importance of resource allocation strategies