The Wall Street Journal reported on September 14, 2026, that OpenAI had acquired Glass Imaging, a company developing image-processing software for smartphone cameras, for more than $300 million. Neither OpenAI nor Glass Imaging has officially confirmed the deal, and TechCrunch noted that OpenAI did not immediately respond to a request for comment. Glass Imaging has disclosed only $31.5 million in outside funding since its founding in 2019 — a full order of magnitude below the reported acquisition price. What's more, the company's technology has an unusual property: its effect grows stronger as camera sensor pixels shrink. OpenAI's first device, reportedly launching in 2027, is said to have no screen and little room for a large lens.
A reported price over $300 million — roughly 9.5 times the funding raised
Outlets don't agree on the exact figure. TechCrunch put "$300 million" in its headline, while SiliconANGLE and the Japanese edition of Investing.com wrote "over $300 million." Since the outlets citing the original report add "over," $300 million appears to be closer to a floor than a fixed number.
Glass Imaging has disclosed three funding rounds. On March 22, 2022, it raised a $2.2 million seed round led by LDV Capital with participation from GroundUP Ventures (per Crunchbase). In February 2024, it raised a $9.3 million extended seed led by GV (Google Ventures), with participation from Future Ventures, Abstract Ventures, and LDV Capital. On May 12, 2025, it officially announced a $20 million Series A led by Insight Partners. Simply adding these up gives $31.5 million. In other words, the reported acquisition price of over $300 million is roughly 9.5 times the $31.5 million Glass Imaging had disclosed in total funding from its founding through the acquisition report. If there were additional undisclosed rounds, the actual cumulative funding could be higher, which would bring the multiple below the 9.5x figure cited here.
The gap looks just as large from the valuation side. Glass Imaging reportedly reached a valuation of about $100 million in its 2025 funding round, and PYMNTS writes that the reported acquisition price is roughly three times that. However, this valuation figure traces back to a single source — the Wall Street Journal report — and both companies remain silent.
In terms of OpenAI's overall spending, this deal is small. Lining up the disclosed figures for OpenAI's hardware-related acquisitions, the $300 million reported for Glass Imaging amounts to about 5% of the $6.5 billion reported for io Products. The calculation 300÷6,500 actually works out to about 4.6%; reaching 5% would require an acquisition price of roughly $325 million. The actual price could be higher than reported, but this hasn't been confirmed.
The price gap is roughly twentyfold, but the two acquisitions target different things. The May 2025 acquisition of io Products gave OpenAI the hardware design organization itself, led by Jony Ive. This time, the target is software that turns signals captured by a camera into images. A deal to buy an entire device-design organization and a deal to buy the software that determines that device's image quality came within 16 months of each other.
Seven years of track record from the two people who built Portrait Mode
Glass Imaging's founders are Ziv Attar, Founder & CEO, and Tom Bishop, Ph.D., Founder & CTO. The company's official press release describes both as "former Apple engineers who created and led the technology behind the iPhone's Portrait Mode." Before that, Attar served as CEO of the Israeli camera company LinX Imaging, which Apple acquired in April 2015 for an estimated $20 million. Apple itself never disclosed the price; the figure is an estimate based on contemporaneous reporting.
Eleven years after Apple's estimated $20 million acquisition of LinX Imaging in April 2015, and seven years after founding Glass Imaging in 2019, Attar's company had built up three rounds of outside funding and a production deployment in a Honor device, culminating in the September 14, 2026 acquisition report. The same person leading a camera company has now moved, over a span of more than a decade, twice between the smartphone-making side and the side trying to build one.
The track record isn't merely theoretical. The company's software, GlassAI, was adopted for zoom processing in Honor's 600 series and shipped in production devices. It was demonstrated at Qualcomm's developer event, Snapdragon Summit, in both 2024 and 2025, showing operation on the Snapdragon 8 Elite and enhanced zoom video. In the May 2025 Series A announcement, Attar said, "GlassAI can unlock the full potential of all cameras to deliver stunning ultra-detailed results." In the same announcement, CTO Bishop explained that the company's image-restoration network exceeds what's possible with other approaches.
What does restoration that works better on smaller pixels actually do?
GlassAI falls into the category of Neural ISP (neural image signal processing). A conventional ISP applies a sequence of separate stages to the RAW data coming off a sensor — demosaicing, noise reduction, sharpening, and multi-frame merging — one after another. GlassAI collapses this into a single network processed in one pass. It also trains per camera module, modeling the individual lens's point spread function (PSF) and the sensor's noise characteristics. Rather than applying the same algorithm uniformly across all devices, it's designed to learn how a specific optical system degrades an image and then reverse that degradation.
This design works best in a specific regime. As pixels shrink, the light captured per pixel decreases, and fine detail gets buried in noise. Conventional ISPs stop gaining resolution in this regime. According to the company's own measurements, as pixel size shrinks from 0.75 micrometers to 0.35 micrometers, neural restoration improved resolution — measured by MTF50, a resolving-power metric — by more than 50%, while conventional ISP performance stayed roughly flat over the same range. This figure comes from Glass Imaging's own measurements, not independent third-party verification.
Shivansh Rao of the company's machine learning team has explained that sub-micron pixels record fine information in ways that conventional processing can't extract. The claim is that the information is captured but not recovered.
Adoption in the Honor 600 series serves as an example of that claim being tested under production conditions. The target is the 200-megapixel main camera shared by the standard and Pro models, using a 1/1.4-inch sensor. The native pixel size is about 0.56 micrometers, and 16 pixels are binned together to function as an effective pixel of about 2.24 micrometers. Binning helps recover light per pixel, but when the center of the sensor is cropped for zoom, the benefit of binning diminishes and the limitations of small pixels show through directly.
Only the standard Honor 600 relies on cropping the main sensor for zoom without a dedicated telephoto lens; the higher-end Honor 600 Pro carries a separate 50-megapixel, 3.5x optical periscope telephoto lens. Restoration matters most precisely where optics can't compensate.
A chassis with no room for a lens, arriving amid litigation with Apple
OpenAI's hardware plans began moving with the May 2025 acquisition of io Products for a reported $6.5 billion. The first device is said to have no screen and to take the form of a movable speaker. According to Bloomberg reporting cited by TechCrunch, it will include a camera and microphone, take a donut shape roughly the size of a hockey puck, cost $300–400, and launch in 2027. All of these details trace back to Bloomberg reporting; OpenAI itself has not disclosed them.
A chassis roughly the size of a hockey puck could have direct implications for the optics. Building a camera into a screenless device of that size leaves little room for a large lens or large sensor, which tends to push pixel sizes smaller. That's exactly the regime where Glass Imaging's technology is said to shine — a fit that aligns with the stated requirements. However, OpenAI has not explained the purpose of this acquisition, and no outlet has yet clearly stated its intended use. The read that the technology's characteristics match the first device's constraints is an inference drawn from placing the two facts side by side — it has not been confirmed as OpenAI's actual intent.
There's another layer of context around the timing. On July 10, 2026, Apple sued OpenAI, its subsidiary io Products, and two former Apple employees in the U.S. District Court for the Northern District of California, alleging misappropriation of trade secrets, among other claims. The named former employees are Tang Yew Tan, OpenAI's chief hardware officer, and engineer Chang Liu. The 41-page complaint alleges the removal of hardware design information, manufacturing methods, and supplier data. OpenAI has denied any interest in competitors' trade secrets, and no legal ruling has been issued.
Amid that ongoing litigation, OpenAI has now brought in a company led by two people who came from the team that built the iPhone's Portrait Mode. Glass Imaging is not named as a defendant in the lawsuit, and no material has surfaced linking the acquisition to the litigation. Still, the pattern remains: people with roots in Apple's camera technology lineage are converging on the OpenAI side.
What comes next: the Honor licensing question and 2027
Whether OpenAI will continue supplying its existing customer, Honor, under license is a particularly hard question to read. It also remains undisclosed what exactly the technology will be used for, and when the two companies might confirm the deal itself. If the buyer locks the technology into its own device, the licensing business would shrink; but because Glass Imaging trains per camera module, insights gathered from running on a variety of production devices could feed back into future designs. No announcement has indicated which direction this will go.
Industry trends point in the same direction as this acquisition. The major chips arriving in late 2026 — Apple's A20 Pro, Google's Tensor G6, and the Qualcomm Snapdragon 8 Elite generation — are all shifting image processing toward the neural engine. The Pixel 11, which carries the Tensor G6, was announced on August 12, and Google hardware executives have said it's manufactured on an improved TSMC 3nm process. Glass Imaging's claim goes a step further than merely offloading part of the pipeline to a network — it replaces the sequential ISP pipeline itself with a network. OpenAI's M&A activity in 2026 has been climbing: Crunchbase reported six deals as of March, followed by the acquisition of TBPN in April and others since. Counting this Glass Imaging deal, the year's total is likely to match or exceed the eight deals recorded for all of 2025.
For readers in Japan, there's a further complication: limited opportunity to verify any of this firsthand. The Honor 600 series, which carries GlassAI, has not been officially launched domestically, making it difficult to compare the technology's real-world performance in stores right now. More evidence will become available in Japan only once OpenAI's first device ships, or once another device carrying GlassAI launches domestically.
The order in which things can be confirmed is set. First, whether both companies acknowledge the deal. Second, whether the handling of supply to Honor is clarified. And third, once the first device — reportedly slated for 2027 — ships, whether third parties can measure how much resolution it manages to extract from a small sensor and short focal length. Only once those measurements exist can outside observers judge whether paying 9.5 times the funding raised was a reasonable price for computationally overcoming optical constraints.