DeepSeek Sparks AI Spending Debate in Tech Giants
NASDAQ:NVDA plunged over 17% Monday as DeepSeek, a Chinese AI company, unveiled an open-source model trained for $5 million with inferior chips, raising concerns about overspending in AI infrastructure among tech behemoths such as NASDAQ:MSFT and NASDAQ:META.
Earnings Calls Shed Light
However, Microsoft and Meta's earnings calls Wednesday provided a different perspective. Microsoft CEO Satya Nadella hailed DeepSeek's efficiency improvements as beneficial to hyperscalers like Microsoft. Meta CEO Mark Zuckerberg downplayed the threat, acknowledging the industry's constant evolution.
Cost Advantage
Analysts questioned DeepSeek's claimed $5 million training cost, citing excluded research and development expenses. However, the use of distillation, a technique for training AI models, highlights potential cost savings in AI training.
Zuckerberg's Perspective
Zuckerberg emphasized that it's premature to adjust AI infrastructure investments based on DeepSeek's announcement. Meta's focus on data centers is driven by the industry shift towards AI models requiring significant computing power.
Nadella's View
Nadella stressed Microsoft's long-term investments in data centers, emphasizing their scalability and customer benefits. Microsoft plans to allocate $80 billion for its AI buildout by fiscal 2025.
Tech Earnings Season
Meta and Microsoft kickstarted this quarter's earnings season, with Apple, Intel, Google, and Amazon set to follow. The latest AI developments are expected to feature prominently in their respective reports.