What is Glass Imaging and why did OpenAI buy it?

Glass Imaging is a Los Altos startup founded by former Apple camera engineers Ziv Attar and Tom Bishop, who worked on the iPhone's Portrait Mode. Its product, GlassAI, is a 'Neural ISP' — an AI image-signal-processing pipeline that corrects lens aberrations, sensor noise and low-light artifacts in real time so small phone cameras produce DSLR-like results; it has shipped in Honor smartphones. The Wall Street Journal reported on September 14, 2026 that OpenAI acquired the company for more than $300 million in a deal completed quietly in recent months. The logic: OpenAI's first consumer hardware, designed with Jony Ive's io team, needs cameras that give a multimodal model clean input, and Glass's team knows how to build that at phone scale.

How does Glass Imaging fit OpenAI's hardware plans?

OpenAI bought Jony Ive's io Products for $6.5 billion in 2025 and has been developing a screen-light, camera-and-microphone-first companion device — a smart speaker-like product is expected first, in 2027, with glasses and other form factors discussed. A device whose main input is what it sees needs a camera stack tuned for AI, not for photos people look at: consistent color, low-light performance and low latency for a model such as GPT-6 Astra to analyze surroundings. Glass Imaging's Neural ISP is exactly that layer, and it can run on the device's own silicon rather than in the cloud.

How much has OpenAI spent on hardware acquisitions?

The two disclosed device-related deals are io Products ($6.5 billion, 2025, all-stock) and Glass Imaging (reported at over $300 million, 2026). OpenAI has also lined up a wave of compute capacity — at least 10 GW of NVIDIA systems, 6 GW of AMD GPUs and 10 GW of custom Broadcom accelerators — and paused new $200 ChatGPT Pro sign-ups on September 10, 2026 because GPT-6 Astra demand outran capacity. Hardware is now a stated pillar of the company's strategy alongside models and ChatGPT.

Will Glass Imaging's technology still ship in Honor or other phones?

OpenAI has not said. Glass had licensed GlassAI to Honor and pitched it to other Android makers. Acquirers in this position typically wind down third-party licensing to keep the technology exclusive to their own device, as Apple did after buying camera and imaging startups. Existing Honor devices keep the software already on them; future licensing is the open question, and neither OpenAI nor Honor has commented as of September 15, 2026.

Quick Answer

OpenAI Buys Glass Imaging for $300M: What It Means (2026)

Published: September 15, 2026

The short answer

OpenAI has bought its camera team. The Wall Street Journal reported on September 14, 2026 that OpenAI acquired Glass Imaging — an AI-camera startup founded by ex-Apple engineers — for more than $300 million, in a deal closed quietly over the preceding months. It is the second hardware acquisition after Jony Ive’s io Products ($6.5 billion, 2025) and the clearest sign yet of what OpenAI’s first device will be built around: seeing.

Who Glass Imaging is


Founded	2019, Los Altos, California
Founders	Ziv Attar (CEO) and Tom Bishop (CTO), both formerly on Apple’s camera team; credited on iPhone Portrait Mode work
Product	GlassAI, a “Neural ISP” that replaces parts of the traditional image-signal-processing pipeline with neural networks
What it does	Corrects lens aberrations and sensor imperfections, denoises low-light frames and recovers detail in real time, so a small lens and sensor produce output closer to a large-sensor camera
Commercial traction	Shipped in Honor smartphones; pitched to other Android OEMs
Prior funding	About $30 million: seed (2021, LDV Capital), $9.3M extended seed (2024, led by GV) and a $20M round (May 2025, Insight Partners)
Deal	Reported at over $300 million; OpenAI and Glass have not disclosed terms

The key technical point: an ISP is the chip pipeline that turns raw sensor data into an image. Glass’s approach trains models per lens-and-sensor combination, which is why it can make a cheap camera module perform above its class — valuable for a first-generation device that needs to be affordable and small.

Why OpenAI wants it

1. The device is camera-first. OpenAI’s collaboration with Jony Ive has produced a companion device concept that is meant to be aware of its surroundings without a screen dominating the interaction. Reporting points to a smart speaker-style product first, expected in 2027, with glasses and other wearables in development. For any of those, the camera is the primary sensor and the model’s eyes.

2. Models need clean input, not pretty photos. Multimodal models such as GPT-6 Astra are sensitive to noise, color shifts and motion blur. A Neural ISP tuned for machine consumption — consistent exposure, low latency, robust low-light — improves what the model can reliably see. Phone makers tune for human viewers; OpenAI can tune for Astra.

3. On-device processing. GlassAI runs on the device’s NPU. That keeps raw frames off the network, cuts latency and reduces cloud compute at a moment when OpenAI is rationing capacity (it paused new $200 ChatGPT Pro sign-ups on September 10 because Astra demand outran GPUs).

4. Talent. Camera-pipeline engineers who have shipped in iPhones and Honor phones are scarce. A $300 million acquihire-plus-IP is cheap next to the io deal.

How it fits the wider OpenAI hardware picture

io Products (2025, $6.5B): Jony Ive’s design studio and team; the device program itself.
Glass Imaging (2026, $300M+): imaging pipeline and camera engineering.
Compute: at least 10 GW of NVIDIA systems, 6 GW from AMD, 10 GW of custom Broadcom accelerators, plus Stargate.
Software surface: ChatGPT’s voice mode, Astra’s computer use and the Agents API give a device something to do once it can see and hear.

Read together, OpenAI is assembling the parts of a consumer electronics company without (yet) announcing a product. Details on the expected device: OpenAI’s first hardware device: smart speaker, price and 2027 release.

How this compares with Apple and Google

Apple integrates its camera stack end to end and, on September 14, 2026, shipped Visual Intelligence as a Siri mode inside the iPhone Camera app. Google’s Gemini Live camera mode runs on Pixel and Android. Both already own the ISP layer OpenAI just bought. OpenAI’s disadvantage is the lack of a phone; its bet is that a dedicated always-on device with a model-first camera can be a better assistant surface than a phone camera app. Glass Imaging narrows the hardware gap; it does not close the distribution gap.

Open questions

Honor licensing. Will GlassAI keep shipping in third-party phones?
Device timing. No launch date; 2027 is the reported target for the first product.
Privacy design. A camera-first home device will face the same scrutiny Amazon’s Echo Look and Meta’s Ray-Ban glasses did. OpenAI has not published a hardware privacy architecture.
Integration with Astra. Whether Glass’s pipeline is used only for capture or also as a preprocessing stage for the model.