Category:
DeepSeek Just Changed Generative Artificial Intelligence (AI) Forever. 2 Surprising Winners From Its Innovation. @themotleyfool #stocks $META $AAPL

DeepSeek Just Changed Generative Artificial Intelligence (AI) Forever. 2 Surprising Winners From Its Innovation. @themotleyfool #stocks $META $AAPL.

DeepSeek’s Revolutionary AI Models: A Game Changer in the Industry

On January 20, the Chinese artificial intelligence start-up DeepSeek made waves in the tech world by unveiling its first-generation reasoning models. The announcement was nothing short of groundbreaking, with the company making some bold claims that have captured the attention of industry experts and competitors alike.

DeepSeek’s DeepSeek-R1 model is said to deliver performance on par with OpenAI-o1, which is widely regarded as the top-performing model across various domains. This achievement is particularly noteworthy given that DeepSeek operates with significantly less advanced hardware compared to OpenAI and other American tech giants. The implications of this are profound, as it suggests that DeepSeek has managed to optimize its resources to an extraordinary degree.

Perhaps even more impressive is the cost efficiency of DeepSeek’s model. The R1 was developed using DeepSeek’s V3 large language model, which was released in December. The company estimates that the compute cost for training V3 was a mere $5.6 million. To put this into perspective, OpenAI’s GPT-4 required a staggering $100 million for training. Achieving similar performance at a fraction of the cost is a testament to DeepSeek’s innovative approach.

Moreover, DeepSeek’s decision to make its model completely open source means that anyone can replicate its techniques, potentially reshaping the entire AI industry. This move could democratize access to advanced AI capabilities, allowing smaller companies and developers to leverage cutting-edge technology without the prohibitive costs typically associated with AI development.

The Long-Term Impact of DeepSeek-V3 and R1

DeepSeek’s focus on maximizing the efficiency of its limited hardware capabilities is a key factor in its success. Due to AI chip export restrictions, Nvidia is unable to sell its most powerful H100 GPUs in China. Instead, it offers H800 GPUs, which are specifically designed to comply with U.S. regulations. These GPUs have a reduced chip-to-chip transfer rate, which slows down the training of large AI models.

In response to these limitations, DeepSeek developed processes to minimize the amount of data that needs to be transferred throughout the system at any given time. For instance, its “mixture of experts,” or DeepSeekMoE, introduced last year, allows only a portion of the model to be activated in response to queries. While other companies also use this method, DeepSeek’s unique approach has made its training process more efficient, reducing the overhead typically required to lower inference costs later on.

Additionally, DeepSeek has developed methods to reduce the memory required for AI inference by compressing important data before storage and transmission. The company has also introduced new approaches to load balancing, which optimizes the distribution of processes across its network of GPUs.

The result of these innovations is an AI model that is not only faster and cheaper to train but also more cost-effective to run. This means that AI inference has become significantly more accessible, even on less-capable hardware. In a world where AI systems can potentially run on devices as small as a smartphone for a fraction of a penny, two major companies stand to benefit significantly: Apple and Meta Platforms.

Making On-Device AI a Reality

Apple has long prioritized data privacy in its development of artificial intelligence features for the iPhone and other devices. Apple Intelligence is designed to operate as much as possible on the device itself, minimizing the need to send data to the cloud. When cloud interaction is necessary, Apple takes extensive measures to encrypt user data during the process.

The new AI features Apple introduced last year are only available on iPhones released in the last 15 months. This is because Apple aims to keep everything on the device, requiring sufficient processing power and memory to run its AI. The latest iPhone chip, the A18 Pro, has increased memory bandwidth to support faster AI processing.

By adopting many of DeepSeek’s methods, Apple could enhance the iPhone’s ability to handle AI inference. This could lead to features such as a more conversational and context-aware Siri, faster translation without an internet connection, smart camera functionalities, and improved productivity tools. These advancements could boost Apple’s iPhone sales and services revenue.

Currently, Apple stock trades at a relatively high multiple of 32.5 times forward earnings. However, with its substantial cash flow, which it uses to buy back shares, and increasing profitability from services revenue, Apple can justify this high multiple. The potential growth from significant improvements to on-device AI could serve as a catalyst for further growth in the coming years.

Scaling AI to 3 Billion People

Meta’s investment in AI is rapidly increasing as it seeks to scale its capabilities and integrate AI features across more areas of its business. Capital expenditures grew by approximately 40% in 2024, with management projecting a 60% increase in 2025. These AI investments have already yielded positive results for Meta, enhancing user engagement, improving advertising tools, and introducing new features like Meta AI, which could become significant revenue sources in the future.

One crucial decision Meta made regarding AI was to open-source its AI model Llama. This decision was partly motivated by the desire to make the model more efficient. In fact, DeepSeek used Llama as the foundation for developing R1, aligning perfectly with Meta’s goals.

Reducing the cost of AI inference could unlock substantial profits for Meta. This has been a longstanding challenge for the company. “A lot of the stuff is expensive, right, to kind of generate an image or a video or a chat interaction,” Zuckerberg said during an earnings call in February 2023. “So one of the big interesting challenges here also is going to be how do we scale this and make this work more efficient so that way, we can bring it to a much larger user base.”

DeepSeek is addressing this challenge and providing Meta with the tools needed to scale AI to its 3 billion users. While Meta may not reduce its AI spending anytime soon, it is now positioned to generate significantly more revenue from its committed investments.

Meta stock has surged on the news of DeepSeek’s innovations, reaching a new all-time high. Despite this, shares are trading at 26.8 times forward earnings estimates as of this writing. Meta is also a cash-rich company, using excess free cash flow to buy back shares and support strong earnings-per-share growth. If Meta can make AI more profitable, it stands to see substantial earnings growth in the coming years, making it a worthwhile investment.

  • DeepSeek’s AI models are open source, allowing widespread adoption.
  • Apple could enhance on-device AI capabilities using DeepSeek’s methods.
  • Meta’s AI investments are poised to yield significant returns.

In conclusion, DeepSeek’s innovative AI models have the potential to revolutionize the industry by making advanced AI capabilities more accessible and cost-effective. As Apple and Meta Platforms stand to benefit significantly from these advancements, the future of AI looks promising and full of potential.

Original source article rewritten by our AI can be read here.
Originally Written by: Adam Levy

Share

Related

Popular

India: AI Journalism Sparks Concern - Credit: Deutsche Welle

bytefeed

By clicking “Accept”, you agree to the use of cookies on your device in accordance with our Privacy and Cookie policies