Glass Imaging, a Los Altos start-up that uses neural networks to pull more detail out of small phone cameras, has been bought by OpenAI in a deal that valued it at more than $300 million, according to a report by The Wall Street Journal on 14 September 2026. The Journal, citing people familiar with the matter, said the acquisition was completed quietly in recent months.

Neither company has confirmed the purchase publicly, and OpenAI did not immediately respond when TechCrunch asked for comment. The Information also reported the deal, under the headline “OpenAI Said to Buy Startup Glass Imaging for $300 Million”. What is not in doubt is who OpenAI would be buying: a team led by two former Apple engineers who worked on the camera technology behind the iPhone’s Portrait Mode, and a product that already ships in Honor’s 600 series phones.

The reported price is roughly three times the $100 million valuation Glass Imaging reached in a funding round last year, and about ten times the roughly $30 million it had raised in total. It lands in the middle of OpenAI’s push into consumer hardware, which began in earnest when the company agreed to pay about $6.5 billion in stock for Jony Ive’s device start-up io Products in May 2025.

This article sets out what the report says, who founded Glass Imaging, how its GlassAI software differs from a conventional camera pipeline, where it is already in use, and why a company best known for ChatGPT might want a camera team. It also separates what is confirmed from what is only reported, because on a deal neither side has announced, that distinction matters.

What the Glass Imaging Report Actually Says

openai buys glass imaging smartphone camera maker 300 million b microchip with a round lens on its raised die

The story rests on one piece of reporting. The Wall Street Journal said OpenAI acquired Glass Imaging, a start-up developing AI-enhanced cameras for smartphones, and that the acquisition valued the company at more than $300 million. Every other outlet covering the deal, including TechCrunch, SiliconANGLE and Israel’s Ctech, attributes the figure to the Journal.

The Wall Street Journal’s account

The Journal’s sources said the deal was “completed quietly in recent months”, which suggests it closed some time before the September report rather than on the day the news broke. The paper described the founders as former Apple employees who worked on camera technology at the company, including its portrait-mode feature.

It also flagged a gap that every follow-up story has repeated: it is not clear how OpenAI intends to use Glass Imaging’s technology. That uncertainty shapes everything else in this article, so the sections on OpenAI’s motives below are analysis, not reporting.

What neither company has said

As of 15 September 2026, OpenAI has not published an announcement, and the Glass Imaging journal page that lists the company’s news carries no entry about the deal. SiliconANGLE noted that neither company had confirmed the purchase publicly.

That leaves practical questions open. We do not know whether the founders have joined OpenAI, which team they would report to, how the price was paid, or what happens to the company’s existing customer agreements.

The Glass Imaging deal at a glance

ItemAs reported
BuyerOpenAI
TargetGlass Imaging, Los Altos, California
Reported valueMore than $300 million (The Wall Street Journal)
Payment structureNot reported
Closing“Completed quietly in recent months”
ConfirmationNone from either company as of 15 September 2026
Previous valuationAbout $100 million, in a 2025 funding round
Capital raisedAbout $30 million in total
FoundersZiv Attar (CEO) and Tom Bishop (CTO)
ProductGlassAI neural image signal processor

Who Is Behind Glass Imaging

openai buys glass imaging smartphone camera maker 300 million c square sensor tile with a grid of tiny square recesses

Glass Imaging was founded in 2019 by Ziv Attar, its chief executive, and Tom Bishop, its chief technology officer. Both came from Apple’s camera work. TechCrunch describes them as the pair who led the team that developed Portrait Mode, the iPhone feature that imitates the shallow depth of field of a large camera lens by separating a subject from its background.

Ziv Attar and the LinX Imaging exit

This is not the first time a technology giant has bought one of Attar’s camera companies. He founded LinX Imaging in Israel in 2009 and ran it as chief executive. LinX built miniature multi-aperture camera modules that combined several small sensors to produce colour images and depth maps.

Apple acquired LinX in April 2015 for a price reported at about $20 million, and Attar joined Apple through that deal. Glass Imaging is, in effect, his second attempt at the same problem: making tiny cameras behave like much larger ones.

Tom Bishop and the Portrait Mode years

Bishop, who holds a PhD, worked alongside Attar at Apple on computational photography. Portrait Mode is a useful clue to what both men know well. It depends on estimating depth from small sensors and then processing the picture so the result looks as if it came from bigger optics.

That instinct, getting big-camera results out of small-camera hardware, runs through almost everything Glass Imaging has published since its founding.

A start-up built around one idea

Glass Imaging’s pitch has been consistent. One of its early technical posts asked whether a $400 smartphone could beat the iPhone 14 Pro’s camera with the help of AI. Rather than adding hardware, the company uses software to extract more of the image quality that a lens and sensor already capture.

It aimed that software at smartphones first, while naming drones and wearables as future platforms in its 2025 funding announcement. For background on how on-device AI is changing handsets more broadly, see our explainer on AI in mobile phones.

How Glass Imaging Raised $30 Million and Tripled Its Price

openai buys glass imaging smartphone camera maker 300 million d small cube in front of a tall slab depth layers

Glass Imaging was never a heavily funded start-up by the standards of AI companies in 2026. Its disclosed rounds add up to roughly $30 million, raised over about four years from a small group of venture investors.

The Glass Imaging funding rounds, in order

DateRoundAmountLeadOther investors
2021SeedNot disclosedLDV CapitalGroundUP Ventures
2024Extended seed$9.3 millionGV (Google Ventures)Future Ventures, Abstract Ventures, LDV Capital
12 May 2025Series A$20 millionInsight PartnersGV, Future Ventures, Abstract Ventures
Reported 14 Sep 2026AcquisitionMore than $300 millionOpenAINone reported

After the Series A, Insight Partners managing director Praveen Akkiraju joined Glass Imaging’s board and Insight’s Jonah Waldman became a board observer. The Journal put the company’s valuation in that 2025 round at about $100 million.

From $100 million to more than $300 million

Measured against its own history, the reported price is a large step up.

Glass Imaging’s money on one scale (US$ millions)
LinX Imaging sale to Apple, 2015 ~$20M
Total raised by Glass Imaging ~$30M
Valuation in the 2025 round ~$100M
Reported OpenAI price $300M+

On those figures the deal values Glass Imaging at about three times its 2025 valuation and about ten times the capital it raised. Because the Journal says “more than” $300 million, both multiples are floors rather than exact figures.

For investors in the $20 million Series A, a sale at roughly triple the company’s valuation no more than 16 months after that round would be a strong result. The actual returns depend on terms that have not been disclosed, such as preference rights and how the price was paid.

What GlassAI Does Differently From a Normal Camera Pipeline

openai buys glass imaging smartphone camera maker 300 million e tall domed speaker cylinder with one round lens

Every digital camera has an image signal processor, or ISP. It is the stage that turns the raw readings from a sensor into a photo a person would recognise. GlassAI is Glass Imaging’s attempt to replace large parts of that stage with a neural network trained for the specific camera it runs on.

How a conventional ISP works

A sensor does not capture colour directly. Most sit behind a colour filter mosaic, so each pixel records only red, green or blue light, and the ISP has to reconstruct full colour for every pixel, a step called demosaicing.

It then reduces noise, sharpens edges, corrects lens shading and distortion, and often fuses several frames into one. In a conventional design these are separate stages, each tuned by engineers, and a later stage can undo detail that an earlier stage recovered.

The Glass Imaging neural ISP approach

SiliconANGLE describes GlassAI as a neural image signal processor that runs raw sensor output through networks trained against a specific camera module. Those networks handle demosaicing, noise reduction, sharpening and frame fusion in a single pass, while modelling the optical behaviour of the individual lens.

Glass Imaging’s own Series A announcement put it more simply: the software reverses lens aberrations and sensor imperfections to extract the full image quality potential of a camera. The company claimed at the time that GlassAI could boost camera performance tenfold.

StageConventional ISPGlassAI neural ISP
Colour reconstructionHand-designed demosaicing algorithmLearned, inside one network
Noise reductionSeparate stage that can smooth away textureLearned together with detail recovery
SharpeningGeneric filters applied around blurCorrects the blur of the specific lens
Lens flawsGeneral shading and distortion correctionModels the optics of the individual module
Multi-frame fusionSeparate alignment and merge stageHandled in the same pass
TuningEngineers tune each stage by handNetwork trained for each camera module
Hardware in demosDedicated ISP siliconThe phone chip’s neural processing unit

Why training on one module matters

Two phones can use the same sensor and still produce different images, because their lenses, filters and assembly tolerances differ. Glass Imaging’s approach is to learn how one particular module blurs and distorts light, then undo exactly that.

The trade-off is effort. Each new camera needs its own training and calibration, which makes the team’s know-how as valuable as the software. If you are new to how networks learn visual features in the first place, our guide to convolutional neural networks covers the building blocks.

The Small-Pixel Problem Glass Imaging Set Out to Solve

openai buys glass imaging smartphone camera maker 300 million f three cylinders rising short medium tall

Phone makers keep adding megapixels without making phones much thicker, so the individual pixels on each sensor keep getting smaller. Smaller pixels collect less light and suffer more from diffraction, the blurring that happens as light passes through a small aperture. Glass Imaging argues that this is where a conventional ISP runs out of road.

Why phone pixels keep shrinking

The Honor 600 is a good example of the pressure. Honor says its 200-megapixel main camera uses a 1/1.4-inch sensor and combines pixels 16-in-1 to reach an equivalent 2.24 µm “super pixel”.

Sixteen-in-one binning groups pixels in a four-by-four grid, so the native pixel pitch works out at 2.24 ÷ 4 = 0.56 µm. That is well under one micron, and it is exactly the kind of sensor Glass Imaging built its software for.

The 0.75 to 0.35 micron comparison

Glass Imaging has published technical results on shrinking pixels. Across pixel sizes falling from 0.75 µm to 0.35 µm, the company says its neural restoration lifted MTF50, a standard measure of how much fine detail a camera resolves, by more than 50 percent, while a conventional ISP largely stalled.

These are Glass Imaging’s own figures, reported by PetaPixel and SiliconANGLE. They are still useful context, because they show why a buyer interested in very small cameras would pay attention.

Putting the pixel sizes in this article on one scale shows how far below the binned “super pixel” the native pixels sit.

Pixel sizes on one scale (micrometres)
Honor 600 binned “super pixel” 2.24 µm
Top of Glass Imaging’s comparison range 0.75 µm
Honor 600 native pixel (2.24 ÷ 4) 0.56 µm
Bottom of Glass Imaging’s comparison range 0.35 µm

Detail recovery, not hallucination

Generative AI can invent convincing texture that was never in the scene, which is a problem for any camera that claims to show reality. Glass Imaging has tackled this in a technical post titled “Restoring Real Smartphone Image Detail, not Hallucinations, with Glass AI”.

Its argument is that a network trained on one module’s physics recovers information the sensor genuinely captured, rather than guessing at plausible detail. That distinction matters even more if the pictures end up feeding an AI system that has to reason about what it sees.

Glass Imaging in Real Phones: Honor 600 and Snapdragon

Unlike many AI imaging start-ups, Glass Imaging has a shipping product. Its software is used for zoom processing in Honor’s 600 series, and it has been demonstrated on Qualcomm’s flagship Snapdragon chips at the chipmaker’s annual summit two years running.

The Honor 600 series

Honor unveiled the Honor 600 and 600 Pro on 22 April 2026, with Malaysia as the first market and Europe following shortly after. The base model has no telephoto camera at all, so its zoom has to come from cropping the 200-megapixel main sensor.

According to PetaPixel’s report on 1 June 2026, GlassAI is applied to the phone’s zoom imaging, “helping to recover fine detail, reduce noise, and preserve natural color and texture across the zoom range”.

SpecificationHonor 600Honor 600 Pro
Main camera200 MP, 1/1.4-inch sensor200 MP, 1/1.4-inch sensor
TelephotoNone50 MP periscope, 3.5x optical
Ultrawide12 MP12 MP
Front camera50 MP50 MP
ChipsetSnapdragon 7 Gen 4Snapdragon 8 Elite
Display6.57-inch AMOLED, 120Hz6.57-inch AMOLED, 120Hz
Battery6,400 mAh (Europe), 7,000 mAh (Asia)6,400 mAh (Europe), 7,000 mAh (Asia)
Malaysia launch priceMYR 2,599 (12GB/512GB)From MYR 3,099 (12GB/256GB)

“We’re honored to partner with Honor and to have our Neural ISP technology featured in the 600 series,” Attar said when the partnership was publicised.

What crop zoom asks of the software

Cropping is simple arithmetic with a hard cost. Doubling the zoom halves both the width and the height of the sensor area used, so a 2x crop keeps a quarter of the pixels: 50 megapixels out of 200.

A 4x crop keeps one sixteenth, or 12.5 megapixels, drawn from native pixels only 0.56 µm across. Recovering clean detail from that small patch is exactly the job GlassAI was built to do.

The megapixels left after each crop fall away quickly, which is why the processing matters more at every step.

Megapixels left after cropping a 200 MP sensor
1x (full sensor) 200 MP
2x crop (200 ÷ 4) 50 MP
3x crop (200 ÷ 9) 22.2 MP
4x crop (200 ÷ 16) 12.5 MP

The Snapdragon Summit demonstrations

Glass Imaging debuted GlassAI at Qualcomm’s Snapdragon Summit in 2024, where Qualcomm presented the software running on its Snapdragon 8 Elite chip. At the 2025 summit on 24 September 2025, the company extended its neural ISP to real-time video.

On a reference device with the Snapdragon 8 Elite Gen 5, it showed 4K video at 20x zoom or higher, running on Qualcomm’s Hexagon neural processing unit. Real-time video is much harder than a single photo, because every frame has to be processed inside the time budget of the recording.

Why OpenAI Wants a Camera Company

OpenAI has not explained the purchase, so any answer is inference. But the deal fits a pattern that has been visible since 2025: OpenAI wants to build its own devices, and it has been assembling the people and technology to do that.

The io Products deal and the Jony Ive device

In May 2025 OpenAI agreed to buy io Products, the hardware start-up co-founded by former Apple design chief Jony Ive, in a stock deal worth about $6.5 billion. Ive and chief executive Sam Altman have since been working on a consumer device.

Public reporting has described that device as pocket-sized, screenless and contextually aware, gathering information about its surroundings through microphones and cameras. OpenAI has not published a specification, so treat those details as reported rather than confirmed.

A device that has to see

If a device is meant to understand its surroundings through a camera, image quality stops being a nicety. A small wearable or pocket gadget has room only for tiny camera modules with small pixels, the exact conditions Glass Imaging has spent years working on.

Better input images should also mean better results from the computer vision models that have to interpret a scene. That gives OpenAI one plausible reason to own camera processing rather than license it from a supplier.

TechCrunch noted that OpenAI has also been rumoured to be developing smartphones, earbuds and AI companion devices. Our earlier look at the OpenAI phone rumour covers what an AI-first handset might change.

Delays, the io name and a 2027 window

The device has not arrived on the timeline OpenAI first suggested. According to a court filing reported by Wired and 9to5Mac in February 2026, OpenAI’s first hardware device will not ship to customers before the end of February 2027, and the company has dropped the io name amid a trademark dispute with the start-up iyO.

If the reported timing is right, a deal that closed in 2026 would give Glass Imaging’s engineers time to influence a camera system before launch. Whether that is the plan is something only OpenAI can confirm.

Where Glass Imaging Fits in OpenAI's Buying Spree

Glass Imaging is small next to OpenAI’s largest deals, but it arrives during a busy period of acquisitions that range from hardware and developer tools to media.

DealAnnounced or reportedReported valueWhat OpenAI gets
io ProductsMay 2025About $6.5 billion in stockJony Ive’s hardware design team
Statsig2 September 2025$1.1 billionProduct experimentation platform; CEO Vijaye Raji became CTO of Applications
TBPNApril 2026Not officially disclosedDaily tech talk show and its audience
Glass ImagingReported 14 September 2026More than $300 millionNeural camera ISP and ex-Apple imaging engineers

What the pattern suggests

By reported price, Glass Imaging is a small fraction of the io Products deal: $6.5 billion divided by $300 million is about 21.7, so io cost roughly twenty times as much. The camera deal looks less like a bet on revenue than the purchase of a specialist team with a working product.

The common thread is capability OpenAI would otherwise have to build slowly: Ive’s team for industrial design, Statsig for testing products at scale, and now Glass Imaging for the camera. Investors will also read it alongside Sam Altman’s comments that an OpenAI IPO will not happen in 2026, which leaves private deals like this one as the main visible signs of strategy.

What Happens to Glass Imaging's Existing Customers

The awkward question in any acquisition of a supplier is what happens to the companies already buying from it. Glass Imaging has licensed its technology to at least one phone maker and worked closely with Qualcomm, and OpenAI does not obviously need to keep serving other hardware makers.

Honor and other Glass Imaging licensees

None of the reports say what the deal means for Honor’s 600 series or for future licensing. Software already installed on phones does not disappear, but future tuning for new camera modules is a different matter, because GlassAI has to be trained for each module.

Any phone maker planning to use Glass Imaging on a future model will want clarity on support, updates and exclusivity before committing. Qualcomm, which has showcased the technology on stage twice, faces a similar question about its demonstrations.

Talent versus technology

Acquisitions like this often come down to people. The reported price buys a team with rare experience across optics, sensors and neural networks, plus years of calibration work on real camera modules.

If OpenAI’s aim is its own devices, the most valuable asset may be the ability to design a camera and its processing together from the start, rather than the licensing business Glass Imaging built.

What Is Confirmed, Reported and Still Unknown

On a deal neither party has announced, it helps to sort the claims by how solid they are.

ClaimStatusSource
Founded in 2019 by Ziv Attar and Tom BishopConfirmedCompany materials and press coverage
$20 million Series A led by Insight PartnersConfirmedGlass Imaging announcement, May 2025
GlassAI used for zoom in Honor 600 seriesConfirmedGlass Imaging and PetaPixel
OpenAI acquired Glass ImagingReported, not confirmedThe Wall Street Journal
Price of more than $300 millionReported, not confirmedThe Wall Street Journal
Deal completed in recent monthsReported, not confirmedThe Wall Street Journal
Payment in cash, stock or bothUnknownNot reported
Founders’ roles at OpenAIUnknownNot reported
Future of Honor and other licencesUnknownNot reported
Use in OpenAI’s first deviceUnknownOpenAI has not commented

The confirmed items are about Glass Imaging’s history and products. Everything about the deal itself still rests on one newspaper’s sources. That is common for an acquisition that closes quietly, but the price and timing should be treated as reported rather than settled until either company comments.

What This Means for Businesses Building With Vision AI

Most organisations will never buy a camera start-up, but the deal highlights something that affects anyone deploying vision AI: the quality of what a model sees can matter as much as the model itself.

Better inputs before bigger models

A model that reads documents, inspects products or monitors equipment works from whatever image the camera delivers. Blur, noise and colour errors at capture limit what any downstream model can recover.

Before paying for a larger model, check whether a better camera, better lighting or better processing would fix the problem more cheaply. Our write-up of LLNL’s camera-based inspection system shows how much careful imaging contributes to accurate measurement.

On-device processing and privacy

In Qualcomm’s demonstrations, GlassAI ran on the phone chip’s neural processing unit rather than in the cloud. Processing images on the device keeps raw footage local, saves bandwidth and avoids network delay, all of which matter for wearables and for any workplace camera that captures people.

Those trade-offs apply to business systems too, and they are cheapest to settle at the design stage. If you are weighing where ML model development and inference should happen, or building an AI strategy that includes cameras, start with where images are captured and who controls them.

Watch supplier risk

The Glass Imaging deal is also a reminder that a specialist supplier can be bought by a company with different priorities. If a product depends on one vendor’s AI component, the contract should cover what happens after a change of ownership.

That means support periods, notice of end-of-life, escrow of models or code where it is proportionate, and the right to keep using the version already deployed. It is far easier to negotiate those terms before a supplier is acquired than after.

Glass Imaging FAQ

Did OpenAI confirm it bought Glass Imaging?

No. As of 15 September 2026 the acquisition had been reported by The Wall Street Journal, citing people familiar with the matter, but neither OpenAI nor Glass Imaging had confirmed it publicly.

How much did OpenAI reportedly pay for Glass Imaging?

The Journal said the deal valued Glass Imaging at more than $300 million. The Information’s headline gave the figure as $300 million. How the price was paid has not been reported.

Who founded Glass Imaging?

Ziv Attar, the chief executive, and Tom Bishop, the chief technology officer, founded the company in 2019. Both worked on camera technology at Apple, including Portrait Mode. Attar previously founded LinX Imaging, which Apple bought in 2015.

What is GlassAI?

GlassAI is the company’s neural image signal processor. It processes raw sensor data with networks trained for a specific camera module, handling colour reconstruction, noise reduction, sharpening and frame fusion in one pass while correcting that module’s lens flaws.

Which phones use Glass Imaging technology?

Honor’s 600 series uses GlassAI for zoom imaging. Glass Imaging has also shown the software on Qualcomm reference hardware, most recently for 4K video at 20x zoom on the Snapdragon 8 Elite Gen 5.

Is OpenAI making a phone?

OpenAI has not announced one. It is developing a consumer device with Jony Ive’s team, which a court filing says will not ship before the end of February 2027, and TechCrunch reports rumours of phones, earbuds and companion devices.

References