About this Resource
<p><i><span style="font-size: 16px;">OpenAI is building its own AI chips as the industry shifts toward open models, custom hardware, and local AI. Here's what this means for NVIDIA and the future of AI.</span></i></p><h2><br></h2><h2>AI News: OpenAI Made a Massive Move Against NVIDIA</h2><p><span style="font-size: 16px;">The AI industry is entering an interesting new phase. For years, NVIDIA has dominated the hardware powering modern AI, but major technology companies are increasingly looking for ways to reduce their dependence on NVIDIA. One of the biggest developments is OpenAI’s move toward building its own AI chips.</span></p><p><br></p><h2>OpenAI Is Building Its Own AI Chip</h2><p><span style="font-size: 16px;">OpenAI has started showing results from its Jalapeno chip, a custom processor designed specifically for AI inference. Inference is the part of AI that happens when you send a prompt to a model and receive an answer. Instead of relying entirely on NVIDIA hardware for this process, OpenAI is developing its own technology to handle more of this workload. Early testing reportedly showed very large performance gains on several open-weight AI models, reaching up to 104× performance in certain tests. The important point is that these chips are currently focused on inference rather than training. OpenAI is still expected to rely heavily on NVIDIA hardware for training its largest models, at least in the near future. Still, the move signals something important: major AI companies want greater control over their hardware instead of depending entirely on one supplier.</span></p><p><br></p><h2>NVIDIA Is Betting on Open-Weight AI</h2><p><span style="font-size: 16px;">While OpenAI is moving toward custom hardware, NVIDIA appears to be taking a different approach. NVIDIA has reportedly agreed to acquire Hugging Face, although the deal had not been publicly confirmed by the companies at the time of the report. Hugging Face is one of the largest platforms for sharing and running open-weight AI models. It can be thought of as something similar to GitHub for AI models. The potential acquisition would make strategic sense for NVIDIA. As more developers and companies use open-weight models, they still need powerful computers and GPUs to run them. By becoming more involved in the infrastructure surrounding open AI models, NVIDIA could continue benefiting even if companies become less dependent on closed AI systems from companies such as OpenAI.</span></p><p><br></p><h2>Open-Weight Models Are Growing</h2><p><span style="font-size: 16px;">There are signs that open-weight models are becoming increasingly important. Data from Vercel's AI Gateway showed a major change in token usage. Two months earlier, closed models represented about 71.6% of token usage, while open models represented 28.4%. More recently, open models accounted for around 62% of token usage, compared with 38% for closed models. However, there is an important distinction: when looking at the number of individual requests rather than tokens, closed models were still ahead. This means open models are growing rapidly, but closed models remain extremely important.</span></p><p><br></p><h2>AI Is Moving Closer to Your Computer</h2><p><span style="font-size: 16px;">Another major trend is the increasing ability to run powerful AI models locally. Apple's new high-end chips, including the M5 Ultra, are designed to provide substantial AI computing power on desktop machines. Configurations with up to 512GB of unified memory could allow users to run very large AI models without sending everything to the cloud. The hardware is expensive, but the direction is significant. Instead of every AI task requiring a remote data center, increasingly powerful personal computers could allow developers and businesses to run sophisticated models locally. That could mean greater privacy, lower ongoing inference costs, and more control over AI systems.</span></p><p><br></p><h2>Open Models Are Getting Surprisingly Powerful</h2><p><span style="font-size: 16px;">The rapid improvement of open-weight models is another important part of this story. GLM 5.3 Flash, for example, has demonstrated performance approaching some leading closed models while being considerably cheaper to operate. Qwen 3.8 Flash is another large open model that shows how quickly the open-model ecosystem is developing. The gap between open and closed models is therefore becoming harder to ignore. Developers no longer have to automatically choose a closed model to get high-quality results.</span></p><p><br></p><h2>What Does This Mean for NVIDIA?</h2><p><span style="font-size: 16px;">OpenAI developing its own inference hardware doesn't mean NVIDIA is suddenly losing its position. NVIDIA remains extremely important for AI training and high-performance computing. However, the industry is becoming more competitive. Companies such as OpenAI, Google and Meta are increasingly developing their own chips or exploring alternatives. At the same time, more powerful open-weight models are creating new demand for AI infrastructure outside traditional closed platforms. This could eventually create a more diverse AI hardware market where companies use a combination of NVIDIA GPUs, custom chips and local hardware.</span></p><p><br></p><h2>The Bigger Picture</h2><p><span style="font-size: 16px;">The most important story isn't simply OpenAI vs. NVIDIA. It is the broader shift toward companies wanting to control more of the AI technology stack. OpenAI is working on custom inference hardware. NVIDIA is strengthening its position in open AI infrastructure. Apple is making powerful local AI computers. Meanwhile, open-weight models are becoming increasingly capable and popular. For everyday users, this could eventually mean more powerful AI, more choices, lower costs and greater ability to run AI privately on personal hardware. The AI race is no longer just about who builds the smartest model. It is increasingly about who controls the models, chips, software and infrastructure needed to run them.</span></p>