Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 📉 $600 billion in market cap loss for Nvidia

• 🐳 How to access DeepSeek R1

• 🤯 DeepSeek R1 has Meta scrambling

• 📝 Perplexity AI proposes TikTok merger

Let’s get right into it!

Sign Up | Advertise | Tools

BIG TECH

$600 billion in market cap loss for Nvidia

Chinese AI lab DeepSeek has reportedly developed a powerful AI model rivaling U.S. counterparts at a fraction of the cost, using older Nvidia chips restricted by U.S. export controls. This breakthrough, achieved with efficient software-driven optimization, has raised concerns among investors about potential disruption to the AI chip market, leading to a significant sell-off in stocks like Nvidia, Broadcom, and AMD.

While some analysts suggest the panic may be overblown, the news highlights the possibility of cheaper AI infrastructure altering global competition and slowing U.S. capital expenditure on AI. Despite skepticism over DeepSeek's claims, analysts maintain that demand for AI infrastructure and high-performance chips remains robust, emphasizing Nvidia's dominance in broader AI applications.

STARTUPS

How to access DeepSeek R1

The DeepSeek AI chatbot, developed by a Chinese startup, has surpassed OpenAI's ChatGPT in popularity on Apple’s U.S. App Store by offering comparable performance for free. Powered by the R1 model, which rivals OpenAI’s o1 reasoning model, DeepSeek demonstrates innovative engineering by using less powerful chips and efficient training methods.

While praised for its accessibility and open weights that allow local use, the chatbot faces challenges common to generative AI, such as inaccuracies, censorship, and opaque data sourcing. Despite these limitations, DeepSeek's cost-effective and high-performing model disrupts the AI landscape, raising questions about the necessity of expensive infrastructure and prompting potential advancements in U.S. AI models.

BIG TECH

DeepSeek R1 has Meta scrambling

Meta has mobilized four "war rooms" of engineers to respond to the release of DeepSeek's R1 AI model, which reportedly rivals leading technologies like ChatGPT at a fraction of the cost. Backed by High-Flyer Capital Management, DeepSeek's breakthrough has raised concerns at Meta, with some insiders suggesting R1 could outperform the next version of Meta’s Llama AI, slated for release in early 2025.

Two of Meta's teams are analyzing how DeepSeek reduced training costs, aiming to apply similar tactics to Llama, while another team investigates DeepSeek's training data and a fourth explores restructuring Llama’s models to compete. As the AI race intensifies, Meta CEO Mark Zuckerberg announced plans to invest $65 billion in AI projects, including data center construction and staffing, amid broader industry moves like the $500 billion Stargate AI infrastructure initiative.

STARTUPS

Perplexity AI proposes TikTok merger

Perplexity AI has proposed a merger with TikTok's U.S. operations, allowing the U.S. government to own up to a 50% non-voting stake in a newly created entity. This revised proposal, submitted after the TikTok ban went into effect on January 19, aims to address national security concerns by ensuring U.S. board control and excluding TikTok's proprietary algorithm from ByteDance's contributions. The plan mirrors strategies previously discussed by former Treasury Secretary Steven Mnuchin, focusing on reducing Chinese ownership while maintaining a U.S.-controlled structure.

With significant interest from investors, including potential discussions involving Oracle, the proposal seeks to strike a balance between satisfying security requirements and preserving ByteDance's ties to TikTok. Meanwhile, President Trump has delayed the enforcement of TikTok's ban for 75 days, expressing optimism about reaching a favorable deal soon, as bipartisan concerns about data security and algorithm control continue to shape the platform's future in the U.S.