OpenAI is reported to be preparing to unveil its first AI device, according to an Axios report. This move could mark a significant expansion beyond software. The device, expected to launch in the second half of 2026, is being developed with former Apple design chief Jony Ive, signaling OpenAI’s ambitions to bring generative AI into a dedicated consumer device.
Best known for developing ChatGPT, OpenAi has undoubtedly changed how people use and perceive artificial intelligence. This generative AI pioneer develops large-scale language, multimodal, and generative models that are applied across customer service, education, and content creation. Through partnerships with major technology and enterprise platforms, OpenAI models have been integrated into everyday tools, expanding real-world impact and adoption of generative systems.
The company reported annualized revenue exceeding $20 billion in 2025, driven by increased computing capacity and enterprise adoption, while prioritizing “practical adoption” of AI in health, science and business sectors in 2026.
To support its rapid growth and infrastructure needs, OpenAI has been actively diversifying its hardware partnerships and supply chain. Earlier this year, OpenAI agreed to purchase up to 750 megawatts of AI computing capacity from chipmaker Cerebras in a deal valued at more than $10 billion, a strategic move that adds to its roster of partners alongside NVIDIA, AMD, and Broadcom.
Alongside hardware expansion, OpenAI has continued to introduce safety-focused features, including age prediction in ChatGPT to strengthen user protections, while also addressing the growing energy demands of its data center footprint, an issue increasingly central to broader industry discussions on sustainability and infrastructure.
OpenAI: Patenting Activity
OpenAI’s patenting activity surged in 2023, reflecting a strategic shift as the company expanded its commercial footprint and intensified its intellectual property efforts. This surge aligns with OpenAI’s move toward commercializing its AI technology, highlighted by major milestones such as the integration of GPT models into Microsoft products.
This increase in filings also corresponds with broader developments in OpenAI’s business and technology roadmap, including partnerships and product milestones. For example, OpenAI has been actively forming major industry collaborations, including a suite of enterprise deals with companies such as Spotify, Zillow, and Mattel, aimed at deepening integration of its AI tools across diverse sectors. Additionally, the company has pursued brand protections and trademark registrations that point to future hardware ambitions, with filings suggesting potential lines of consumer devices and robotics.
Beyond patents and trademarks, OpenAI’s overall strategy reflects a mix of defensive and commercial positioning. Legal activity around intellectual property has also featured in the news, including trademark disputes over the “Open AI” name and contested partnerships, such as a halted hardware collaboration with designer Jony Ive due to legal challenges. At the same time, broader industry context shows that generative AI patenting has expanded rapidly worldwide, with firms in China filing significantly more AI patents overall, while OpenAI only began substantial filings in 2023 after years as a research-centered organization. These developments illustrate how OpenAI’s surge in patent activity fits within a larger narrative of rapid growth, competitive pressures, and strategic legal positioning in the AI landscape.
OpenAI: Top Technology Areas
OpenAI’s innovation is heavily concentrated in digital data processing and advanced computational systems, particularly in areas that support large-scale AI model training and deployment.
The strong emphasis on G06F reflects a focus on the underlying computing infrastructure that enables efficient execution, optimization, and scaling of complex models. Alongside this, G06N highlights the central role of machine learning architectures, neural networks, and specialized AI computation methods in OpenAI’s technology stack. Together, these domains show that OpenAI’s core strength lies in combining powerful AI models with robust system-level engineering, ensuring performance, reliability, and scalability.
Beyond these primary areas, OpenAI’s work extends into vision technologies, networking, and hardware-related innovations. Classifications such as G06V and G06T indicate continued development in image and video understanding and generation, supporting multimodal capabilities. H04L points to communication and data transmission technologies that enable distributed AI systems, while G11C and H10B suggest attention to memory, storage, and hardware efficiency.
OpenAI Patents
OpenAI’s featured patents showcase the technologies behind its advances in language models and generative AI. They highlight how the company transforms research into practical tools that improve reasoning, creativity, and real-world problem solving across industries.
Flexible, accurate, and guided AI image generation
Traditional image generation systems try to create pictures from text by learning links between words and images. However, these systems often produce blurry, low-quality, or unrealistic images. Sometimes the images do not match the text at all or are confusing and hard to understand. Most systems can only make one image per text and cannot easily make changes while keeping important details intact.
OpenAI’s system solves these problems using a multi-step approach. First, the text is converted into a digital representation. A first model uses this to create a preliminary image representation, and a second model produces the final image, improving quality and resolution with advanced upscaling techniques. The system trains text and image models together so that the images better match the descriptions. It can generate variations, modify images based on additional input, and produce images in different styles or with finer details. Users can guide the system with extra text and access intermediate steps to refine the results.
U.S. Patent No. 11,922,550, titled “Systems and methods for hierarchical text-conditional image generation”, was filed on March 30, 2023, and was granted on March 5, 2024. The patent lists Aditya Ramesh, Prafulla Dhariwal, Alexander Nichol, Casey Chu, and Mark Chen as inventors.
Building on its text-to-image technology, OpenAI uses the same step-by-step approach to edit and expand existing pictures, making it easier to improve, change, or enlarge images while keeping them realistic and detailed.
From pixels to perfection with next-generation AI
Creating images with AI is hard because computers have trouble getting the details right, like lighting, texture, perspective, and style. Making high-quality images takes a lot of computing power, and changing or enlarging images can easily distort important parts.
With this new approach, OpenAI allows users to select a part of an image to change or expand, and the system masks that area. A machine learning model then uses the original image, the masked region, and a text description from the user to generate the updated image. The model can replicate existing pixels, add new details, or create extensions beyond the original image boundaries while keeping the style and meaning consistent.
The system includes sub-models for creating intermediate representations and final outputs, and it allows users to provide input through a graphical interface, making it easier to guide the changes.
U.S. Patent No. 11,983,806, titled “Systems and methods for image generation with machine learning models”, was filed on August 30, 2023, and was granted on May 14, 2024. The patent lists Aditya Ramesh, Alexander Nichol, and Prafulla Dhariwal as inventors.
Unified transformer models for accurate speech recognition
Current speech recognition systems are complicated, with parts for detecting speech, identifying speakers, and converting numbers or abbreviations. They often struggle with accuracy, especially for multiple languages or tasks like transcription, translation, or timestamps. While pre-training audio models has helped, most systems still can’t reliably turn speech into text in different situations without a lot of retraining.
OpenAI solves these problems with a unified approach using a transformer model that includes both an encoder and a decoder. This model can handle multiple languages and multiple tasks at the same time. It can transcribe audio, translate speech into other languages, generate timestamps, and normalize text automatically.
Users can provide audio input, and the system can process it in segments, predicting timestamps, detecting when no speech occurs, and generating accurate text. The system uses tokens to specify the language and task, allowing flexible output for transcription, translation, or timing. By combining all these steps into one model, it reduces complexity, improves reliability, and works effectively across different languages and scenarios without requiring separate models or extensive fine-tuning.
U.S. Patent No. 12,079,587, titled “Multi-task automatic speech recognition system”, was filed on April 18, 2023, and was granted on September 3, 2024. The patent lists Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey-Payne, and Ilya Sutskever as inventors.
On top of these capabilities, OpenAI has developed patented technologies that power ChatGPT, their well-known AI product, enabling more intuitive image interaction and context-aware text editing.
Reimagining user interaction with images
Traditional ways of interacting with images in a graphical interface can be slow and difficult. Users often have to move a cursor to select specific parts of an image, which can be tedious and not precise.
OpenAI’s system addresses these challenges using a multimodal machine learning model that understands both images and text. Users provide a textual prompt about an image, and the model identifies the exact location of the element. The system can highlight the location, move the cursor there, or display a graphic or textual explanation at that spot. It can handle multiple locations simultaneously, show sequential highlights, or modify the image to emphasize selected areas. This approach improves accessibility, accuracy, and overall user experience, allowing users to interact with images more intuitively and efficiently.
U.S. Patent No. 12,051,205, titled “Systems and methods for interacting with a large language model”, was filed on September 27, 2023, and was granted on July 30, 2024. The patent lists Noah Deutsch and Benjamin Zweig as inventors.
Analyzing context for smarter outputs
Conventional language models often struggle to understand natural language prompts and make precise text or code edits, especially across different tasks or when integrated with other systems. They may also need manual adjustments or specific datasets, limiting their flexibility and efficiency.
OpenAI solves these limitations with a system that combines a data input engine, normalization engine, context analysis engine, and language model access engine to handle user input, instructions, and model parameters. The data input engine collects user input, which can include text, code, a prompt, or even a null set, along with instructions that define tasks or constraints like tone, structure, or format.
The system then normalizes, tokenizes, and analyzes the input, while the context analysis engine extracts details such as location, person, time, event, or cause. The language model access engine uses these parameters to generate output and can optimize the model using reinforcement learning or other machine learning techniques. It can adjust the model, use training or demonstration datasets, and refine outputs based on user-labeled metrics. Iterative editing allows the system to retain context across multiple rounds, producing accurate, flexible, and context-aware text or code that improves over time with user feedback and training data.
U.S. Patent No. 11,983,488, titled “Systems and methods for language model-based text editing”, was filed on March 14, 2023, and was granted on May 14, 2024. The patent lists Raul Puri, Qiming Yuan, Alexander Paino, Nikolas Tezak and Nicholas Ryder as inventors.
All featured patents were represented by a team of attorneys from Finnegan, Henderson, Farabow, Garrett & Dunner LLP.
OpenAI: Top Law Firms
OpenAI handles most of its patent work with a few key law firms. Finnegan, Henderson, Farabow, Garrett & Dunner LLP manages the majority of its patent filings, with Yelena Morozova serving as the lead attorney overseeing many of OpenAI’s core filings. .
Besides Finnegan, OpenAI also works with other firms such as Van Pelt, Yi & James and Polsinelli, with Brian McKnight serving as key contact for specialized filings. These firms handle smaller portions of OpenAI’s patent portfolio, likely focusing on specific cases or specialized areas.
