Why Sam Altman Is Steering OpenAI Away From Smart Glasses—and Defending AI Data Centers

Sam Altman’s recent comments point to two connected ideas about AI’s physical footprint: the technology has to be socially acceptable when people wear it, and infrastructure claims need to be evaluated locally rather than through broad assumptions.

Why Altman is skeptical of smart glasses

Altman said he finds it “very uncomfortable” to talk to someone wearing “a camera and a light.” The remark is a clear criticism of the social dynamic created by camera-equipped smart glasses, including products such as Meta’s Ray-Ban glasses. It does not amount to a binding corporate promise that OpenAI will never make glasses, but it makes the category an unlikely near-term priority based on his public comments. 211

The objection is less about whether glasses can run AI features than about consent and comfort. A visible indicator light may signal recording, but it does not remove the uncertainty another person can feel when facing a camera built into someone’s eyewear. That tension has become part of the public backlash around camera-enabled wearables. 2

What OpenAI says it is exploring with Jony Ive

Altman has described a broader family of AI devices developed with Jony Ive’s team:

• Tabletop or desk devices for a fixed setting
• Pocketable devices for portable use
• Body-worn devices for some future use cases 69

That is a strategy statement, not a product roadmap. The most detailed reporting on the first OpenAI device has described it as compact, screen-free, context-aware and not eyewear—a possible “third core device” alongside a laptop and phone that could rest on a desk or fit in a pocket. 114

There is an important distinction between the broader device family and the first product. Court-filing reporting said the first device would not be an in-ear product or a wearable, which means Altman’s comments about body-worn hardware should not be read as confirmation that the debut device will be worn. 38

When could OpenAI’s first consumer device arrive?

OpenAI has not publicly confirmed a release date or final form factor. Early reporting tied the project to a 2026 target, while later reporting described a possible shipment as early as the following year. The reports therefore point to a product still in development rather than a dependable launch window. 110

The practical takeaway: expect more hardware experimentation, but do not treat any particular device category, feature set or date as settled until OpenAI announces it.

Altman’s response to data-center water concerns

In the same broader debate over AI’s real-world effects, Altman pushed back on the idea that modern data centers are inherently enormous water users. He called that belief a persistent “meme” and cited a 2024 Virginia review that, according to reports of his comments, found that most individual data-center buildings used roughly as much water as—or less than—a large office building, though some facilities used substantially more. 4751

He also used a deliberately striking comparison: one almond, he said, can require the water equivalent of about 38,000 ChatGPT queries. The comparison is meant to challenge exaggerated per-query claims, not to measure the full local impact of constructing and operating AI infrastructure. 50

Why local opposition has not gone away

A low average per-query estimate cannot answer whether a particular facility is suitable for a particular watershed, power grid or community. In a University of Virginia survey reported in August, more than three in five Americans opposed a new data center in their own community, while fewer than one in five supported one. Water use, power demand, environmental effects and the visibility of developer agreements were recurring concerns. 1819

Virginia illustrates why the issue is location-dependent. A later state assessment reportedly found that no evaluated location in the Coastal Plain could support a major new groundwater withdrawal under current permitting standards. 20 That finding does not establish that every data center is water-intensive, but it does show why regional supply constraints matter—especially for projects relying on large new withdrawals.

The bigger signal for AI product strategy

Altman’s positions are consistent with a simple test for AI’s next phase: products and infrastructure must work not only technically, but also around other people. For hardware, that means avoiding form factors that make bystanders feel surveilled. For data centers, it means moving beyond global averages and addressing the water, power and disclosure questions facing each host community.

OpenAI’s upcoming devices remain largely unknown. What is clearer is the company’s stated direction: a range of AI-first hardware beyond the phone and glasses, paired with a public argument that AI’s environmental costs should be judged with more precision than viral comparisons allow. 651