Sanctions have prevented DeepSeek from purchasing the NVIDIA GPU, which is necessary to train the AI model that is as powerful as Openai’s Chatgpt O1 inference model. The Chinese startup, which was unable to purchase the required AI hardware, devised another way to train Deepseek R1 inference models and sent shock wave to the world.
The cost of DeepSeek R1 training is 3 % to 5 % of Chatgpt O1 training costs. DeepSeek models are also cheaper and reduces access costs. In addition, since the company has created an AI open source, you can install DeepSeek on your computer and run it locally. Since the training dataset and instructions are still secret, it is at least a commercial product.
These developments were the most affected by NVIDIA, and took the market. Suddenly, investors have noticed that AI companies like Openai do not need to accumulate more computing power to develop a better version of AI.
But there is one better stock than the market, and that is Apple. Considering that Apple Intelligence is now the present than Chatgpt O1, operator, Gemini, Deepseek R1, etc., it may seem surprising development.
However, Apple has a unique approach to AI, and Deepseek’s innovation can help it to provide the future of the AI that you want to provide to your iPhone user. And Apple does not propose Deepseek to replace Apple Intelligence ChatGpt. Instead, Apple may learn from Deepseek’s innovation and copy them.
While the market was falling freely on Monday, I was inappropriate about NVIDIA GPU hardware suddenly becoming outdated. Yes, DeepSeek may have come up with a more efficient way to train AI as smart and abilities as Chatgpt. But that doesn’t mean you don’t need access to high -speed and reliable AI hardware.
Perhaps the fact that the registration of DeepSeek is temporarily restricted for cyber attacks tells you that you can explain another explanation. DeepSeek’s infrastructure may be restricted to respond to demand. Cyber attacking all of it sounds much better than acknowledging that AI requires a lot of power to get off the ground.
That’s all speculation, but time will quickly answer the mystery. Witnesses cyber attacks, registration resuming, or long -term restrictions that show other problems.
I also said on Monday that China, which exceeds US AI companies, is temporary. Innovation introduced by DeepSeek is reproduced throughout the industry. They were probably already so. What happens if entities like Openai and Google use AI training similar to DeepSeek? You can see even faster innovations.
Again, it’s speculation. But everyone will copy everyone in the technology.
So how does this bring profits to the iPhone’s Apple Intelligence? Let’s start from the basics.
Remember that Apple is the only high -tech company that announced a large AI project with a privacy in the core. Apple Intelligence is mainly executing devices. If that is not possible, Apple Intelligency is called Apple’s private cloud computing and moves information to the Apple server.
Apple’s iOS 18.4 update provides a large Siri upgrade seen on the WWDC last year. Siri can analyze on -devices that store more user data and provide even better assistants to iPhone users. The problem of this Siri is that it is not a chatbot. Since Apple has no replacement for Chatgpt, we have built Chatgpt access to Apple Intelligation. Siri chatbots may come to iOS 19 next year.
When Apple is ready to provide chatbots similar to Chatgpt O1 or Deepseek R1, you need to find a way to run on your iPhone. Therefore, Deepseek Tech may be convenient. Especially the distillation process. Ben Thompson explained all of it in Deep Shik FAQ. This refers to training a smaller model using a bleeding edge AI model or model.
Distillation is a way to extract understanding from another model. You can send the input to the teacher model, record the output, and use it to train the student model. This is a way to get a model like a GPT-4 turbo of GPT-4. Companies have a complete access, so it is easier for companies to do their own model, but they can be distilled in a way that is somewhat difficult to handle through the API.
Distillation obviously violates the terms of use of various models, but the only way to stop it is to actually block access through IP prohibition, rate restrictions, etc. It is assumed that it is widely used for model training. GPT-4O The number of models that converge to quality is an increasing number of models. This does not mean that DeepSeek has distilled 4O or Claude, but it is strange to say frankly.
Apple uses this technology to train a special Apple Intelligence model running on the iPhone. Consider the “Siri mini” AI model that only processes conversation dialogue through the text and audio on the iPhone. Another mini -model may be used for other specific tasks on the iPhone so that these tasks are executed on the iPhone.
This will receive AI inference and user commands to provide cheaper, faster, faster, faster, and more private answers than other devices. Thompson identifies a great winner after the research of Deepseek R1, and Apple is one of them.
Apple is also a big winner. Dramatically reducing inference memory requirements will make the edge a lot of execution, and Apple has the best hardware for it. Apple Silicon uses a unified memory. In other words, CPU, GPU, and NPU (neural processing unit) can access shared memory pours. This means that Apple’s high -end hardware actually has the best consumer chip for inference (NVIDIA GAMING GPUS is the largest in VRAM 32GB and Apple chips are 192 GB RAM. Masu).
There is also the fact that DeepSeek has done what we have learned for many years. This optimizes software executed with more hardware. The iPhone did not match Android in terms of specifications, but led the market with a high -end A series chip. Apple optimized iOS Experience, executed with a more limited amount of RAM, and provided a high -speed mobile experience that does not affect the battery life.
DeepSeek has achieved the same thing with AI. Using software optimization, I used AI hardware with lower ability than Openai to train Chatgpt O1 rivals. Everyone is interested in duplicating it. In particular, it is a company that can access the latest NVIDIA hardware.
Apple is likely to pay attention to all of these developments, and may see results in the near future. Of course, I guess, who can ignore DeepSeek’s AI innovation in their right heart? In particular, when you are at the core of all products created by AI.
Finally, DeepSeek points out that this week has created news for topping the App Store and has changed the iPhone to a reliable device to sample new AI innovations that have not been tied to Apple Intelligence. Unlike Apple Intelligence, DeepSeek works on the current iPhone, like the Chatgpt standalone app.
