All eyes were on Nvidia on Wednesday, February 26, as the American chipmaker reported its first quarterly earnings after last month’s panic over DeepSeek. Wall Street analysts and investors closely watched the results for any potential signs that tech giants such as Google, Oracle, Amazon and Microsoft — who buy up to half of Nvidia’s AI chips — might scale back spending. However, Nvidia CEO Jensen Huang has a reason to remain bullish, as the company not only beat Wall Street expectations but also exceeded its own estimates with strong fourth-quarter earnings. Nvidia reported that its revenue for the quarter reached a record-breaking $39.33 billion, indicating that the company’s historic growth run driven by the AI boom will continue into 2025. Since the emergence of generative AI, the company has had a virtual monopoly over its highly-prized chips called graphics processing units (GPUs), which crunch data for AI models and drive AI applications. During the fourth-quarter of fiscal 2024, Nvidia’s net income rose to $22.09 billion and its sales more than tripled. Its full fiscal-year revenue surged 114 per cent to $130.5 billion. "In all, the world’s leading chipmaker, based in Santa Clara, California, has once again delivered strong results, but with the AI sector evolving rapidly and macroeconomic uncertainties looming, the margin for error is razor-thin. Investor sentiment remains cautious, as any perceived weakness — be it in financials, competition, or macro conditions— could weigh on the stock," Subho Moulik, the founder and CEO of online trading and investment platform Appreciate, told indianexpress.com. Nvidia’s AI chip sales Sales of Nvidia’s flagship Blackwell chips as well as the previous generation Hopper chips are reported as part of the company’s data centre business. Over 91 per cent of Nvidia’s total sales come from its data centre sales, which nearly doubled in 2024 to $115 billion and rose 16 per cent from the previous quarter, as per the earnings report. The company reported $11 billion revenue from its Blackwell chips during the fourth quarter. “Blackwell sales were led by large cloud service providers which represented approximately 50 per cent of our Data Center revenue,” chief financial officer Colette Kress was quoted as saying by CNBC. Data centre revenue in the fourth quarter was up 93 per cent on an annual basis at $35.6 billion. Its total data centre revenue has grown 10x over the past two years. Betting on Blackwell chips Nvidia projected its first-quarter revenue in FY 2025 to reach $43 billion. The company’s growth expectations depend on how soon its next generation of Blackwell AI processors can be shipped. CFO Kress said that the firm expects “a significant ramp up” of sales of its next-generation Blackwell AI chips in the first quarter. Meanwhile, CEO Huang termed the demand for Blackwell as “amazing”. While Nvidia’s chips were initially used to develop and train AI models, the company’s executives told investors that they expect the new Blackwell chips to be used for inferencing. Inferencing is a crucial post-training technique, where new data is generated from a trained AI model to observe the patterns it has learned and stored in its parameters. DeepSeek-R1: The elephant in the room Chinese AI startup DeepSeek shot to fame in January this year, after the release of its open-source AI reasoning model R1. DeepSeek's AI chatbot app temporarily dethroned OpenAI’s ChatGPT as the most downloaded app on Apple’s App Store in the US. These developments triggered a broader sell-off of tech stocks across markets, with Nvidia stock falling more than 15 per cent (the largest single-day decline for a public company). Nearly $600 billion was wiped off from the American chipmaker’s market value. The markets were primarily shaken by the realisation that it was possible to develop cutting-edge AI models rivaling OpenAI’s o1 model on thinner margins and with resources that are lower in quality than those used by US counterparts. However, tech leaders such as Nvidia’s Huang and Microsoft CEO Satya Nadella have repeatedly held that DeepSeek’s purported breakthrough will exponentially increase demand for AI computing capacity. During the earnings call with investors, Huang said, “DeepSeek-R1 has ignited global enthusiasm. It’s an excellent innovation, but even more importantly, it has open sourced a world-class reasoning AI model. Nearly every AI developer is applying R1.” He further emphasised that reasoning models such as R1 are good news for Nvidia as they need much more compute power. “Reasoning models can consume 100 times more compute, and future reasoning models will consume much more compute,” Huang said. Kress also looked to address investor concerns over cost-efficient AI models such as DeepSeek-R1. “New ways of running AI models that ask the AI to generate additional information to “think” through responses could require as much as 100 times the amount of Nvidia chips,” she said. “Long-thinking, reasoning AI can require 100 times more compute per task compared to one shot inferences,” Kress added. “The vast majority of our compute today is actually inference,” Huang told investors. There have been other signs that indicate the market for AI chips is not going to cool off in the post-DeepSeek era. For instance, big tech companies such as Meta, Google, and Amazon have all announced investments that are collectively worth hundreds of billions of dollars, for setting up their own AI infrastructure in the coming years. Nvidia’s Huang on competition In response to investor questions about whether Nvidia sees the custom chips in development by Amazon, Microsoft, and Google as a threat to its business, Huang said, “Just because the chip is designed doesn’t mean it gets deployed.” However, the chip giant still sees Chinese telecom major Huawei as a competitor. Huawei was added to Nvidia’s list of current competitors for a second straight year, as per a new filing. This comes despite years of US government restrictions on Huawei. Since 2019, the White House has restricted Huawei’s access to advanced 5G chips and Google’s Android operating system, as the company has been blacklisted citing national security. Nvidia listed Huawei among its competitors in four of five categories, including chips, cloud services, computing processing and networking products. “There’s a fair amount of competition in China. Huawei, other companies, are quite vigorous and very, very competitive,” Huang was quoted as saying by CNBC. Nvidia’s other businesses Nvidia recorded $3 billion in sales from its networking components used to connect stacks of GPUs together - down nine per cent from the previous year. The company reported $2.5 billion in sales from its graphics processors for playing 3D games - a 11 per cent decline on an annual basis. The company had announced its latest line of GPUs called the RTX-50 series at CES 2025 in January this year. The graphic cards run on the same flagship Blackwell architecture, as opposed to the RTX 40-series which shared its Ada Lovelace architecture. But weeks after its launch, some RTX 5090 buyers complained of melting power connectors. Nvidia responded that the problem was due to the connectors not being inserted into the stock on the GPU or being bent in a way that puts stress on the cable itself. Besides graphic cards for gaming, Nvidia also sells chips to power cars and robots. It is said to be one of the company’s growth areas as the company reported $570 million in automotive sales during the quarter - a 103 per cent increase year-over-year (YoY).
