Subscriptions Are Not Enough. Why OpenAI is Betting Everything on Hardware.
By Creatives Takeover · July 29, 2026
OpenAI’s push beyond subscriptions into hardware.
OpenAI generates roughly $2 billion a month in revenue. It also loses $1.22 for every dollar it earns. In 2025 the company spent approximately $22 billion against $13 billion in sales. For 2026, internal projections point to a $14 billion net loss on an annualized revenue run rate of roughly $25 billion, a burn rate of 55 to 65 percent of every dollar that comes in the door. Cumulative losses through 2029 are internally projected at approximately $115 billion, with profitability not expected until sometime between 2029 and 2031, a timeline that has already shifted more than once.
Those are not the financials of a company casually experimenting with a side project. They are the financials of a company under real pressure to find a business model that eventually produces more cash than it consumes. ChatGPT subscriptions and API access, the two revenue lines currently carrying the entire business, are growing quickly but are not growing fast enough, on their own, to close a gap of that size on the timeline OpenAI has committed to. Against that backdrop, the company's most closely watched strategic bet is not a new model release. It is a physical device.
The $6.4 Billion Acquisition That Signaled the Pivot
In May 2025, OpenAI acquired io, the AI hardware startup founded by legendary Apple designer Jony Ive, in an all-equity deal valued at $6.4 billion. The acquisition brought roughly 55 employees in-house, including io co-founders Scott Cannon, Evans Hankey, and Tang Tan, all Apple veterans who had previously worked on the iPhone and other flagship hardware. Ive himself took on what OpenAI described as deep creative and design responsibilities across both OpenAI and io, while keeping his independent design firm, LoveFrom, as a separate entity.
Sam Altman's framing of the deal at the time was not modest. He described the io acquisition as a decision that could deliver an additional one trillion dollars in value to OpenAI, a company that, by his own internal projections, is not expected to turn a profit until 2029 at the earliest. That is a striking gap between the scale of the ambition being attached to a single hardware device and the scale of the losses the core business is currently absorbing every single quarter.
What the Device Actually Is
The product itself, as described in reporting from Bloomberg, the Financial Times, and Axios, is a screen-free, voice-first companion device, internally discussed as something closer to "calm computing," an object designed to reduce screen dependency rather than compete head-on with a smartphone. It is being positioned not as another app or gadget, but as a new category of home computer for the AI era: something that can control smart home devices, play media, answer questions, and hold a natural, ongoing conversation, drawing on the full range of ChatGPT's capabilities without a screen mediating the interaction.
OpenAI's Chief Global Affairs Officer Chris Lehane confirmed at Davos in January 2026 that the company intended to debut its first consumer hardware device in the second half of the year, with initial production targets of 40 to 50 million units through manufacturing partner Foxconn, following an earlier shift away from Luxshare, the China-based contractor best known for assembling iPhones. Altman described seeing the first working prototypes in November 2025 in visibly enthusiastic terms, saying he could not believe how good the work was.
The Delays That Followed
The timeline has not held. In February 2026, court filings related to a trademark infringement lawsuit filed by audio device startup iyO revealed that OpenAI's first hardware device would not ship to customers before the end of February 2027, a shift of at least several months, and potentially closer to a full year, from the originally stated goal of shipping before the end of 2026. The filings also confirmed that OpenAI had dropped the "io" name entirely from any future hardware branding to resolve the trademark dispute, and, notably, that the company had not yet produced any packaging or marketing materials for the device, a detail that suggests the delay is not merely cosmetic.
The technical obstacles behind the delay are, according to Financial Times reporting, genuinely difficult ones rather than simple scheduling issues. Engineers are still working through how to give the device a coherent personality, reportedly aiming for something its designers have described internally as "a friend who's a computer" rather than anything resembling a romantic AI companion. They are also navigating serious privacy questions inherent to a device designed to be always listening in someone's home, and grappling with the physical constraints of power, heat, and bandwidth that come with running always-on AI inference on a small, battery-powered object, constraints that, as of the most recent reporting, do not yet have a clean engineering solution.
Those are not trivial problems, and the industry has a recent, cautionary case study close at hand. Humane's AI Pin and the Rabbit R1, two earlier attempts at a screen-free, always-on AI hardware category, both launched to significant fanfare and both failed commercially soon after, undone by exactly this combination of unclear value proposition, unresolved privacy concerns, and hardware that could not comfortably support the compute the product promised. OpenAI and Ive are reportedly taking the extra time specifically to avoid repeating that outcome, but the extra time itself is also a tacit admission that the problems Humane and Rabbit ran into have not yet been definitively solved by anyone, including a team stacked with some of the most accomplished hardware designers in the industry's history.
Why Hardware, Specifically, Is the Bet
The strategic logic behind reaching for a physical device, rather than doubling down further on software alone, becomes clearer when set against OpenAI's underlying cost structure. Inference costs, the ongoing computational expense of actually running ChatGPT for every user query, are projected to reach $14.1 billion in 2026 alone. That is a cost that scales directly with usage, and a subscription or API pricing model, however well optimized, is fundamentally a battle to keep revenue growing just ahead of a cost base that grows in lockstep with it. There is a structural ceiling to how much margin that dynamic can ever produce.
Hardware offers a different economic shape entirely, at least in theory. A device sold at a healthy markup, potentially bundled with a premium subscription tier, creates a revenue stream with a fundamentally different cost profile than pure API inference: one-time (or infrequent) manufacturing and distribution costs against a durable, owned piece of hardware sitting in a customer's home, generating recurring engagement and, ideally, recurring subscription revenue on top of the device itself. It is, in effect, an attempt to borrow Apple's own business model, hardware margin plus a services layer riding on top of it, for a company that has, until now, operated entirely as a software and API business.
There is also a defensive logic sitting underneath the offensive one. Google, Amazon, and Apple all already control a hardware layer, phones, speakers, watches, that sits physically closer to the end user than any chatbot app ever can. If AI interaction increasingly moves toward ambient, voice-first, always-present computing rather than a screen someone deliberately opens an app on, a company with no hardware presence risks becoming a backend intelligence layer that a hardware-owning competitor can route around, replace, or deprioritize at will. Owning the device is, in that reading, less about growth and more about survival: securing a direct, undisintermediated relationship with the end user before someone else's hardware does it first.
The Case for Genuine Skepticism
None of this should be read as a confident prediction that the bet will work, and the skepticism circulating around it is grounded in more than just habitual cynicism about hardware launches.
The device is entering a category, screen-free always-on AI companions, that has already produced two well-funded, well-designed, high-profile commercial failures in Humane and Rabbit, both of which struggled with precisely the problems OpenAI's own team is reportedly still working through: unclear day-to-day utility relative to a smartphone people already carry everywhere, unresolved privacy concerns around always-on listening, and the physical limits of running meaningful AI inference on a small, battery-constrained device without either compromising the experience or bleeding compute costs the moment the product actually finds an audience. A one trillion dollar value projection attached to a category with that track record, from a company that has not yet finalized the product's basic personality or resolved its power and privacy architecture, is, at minimum, a projection built on considerable unresolved uncertainty.
There is also a sequencing risk specific to OpenAI's financial position. The company is burning through cash at a rate that has already forced it to seek additional capital well beyond its existing commitments, with HSBC estimating OpenAI will need to raise at least $207 billion by 2030 simply to sustain its currently planned rate of loss. Launching an unproven hardware category, into a market with two recent high-profile failures, while the core software business is still several years from profitability, is a genuinely high-variance bet: the kind of move that can either meaningfully diversify a fragile revenue base or meaningfully compound an already difficult cash position, depending entirely on execution the company has not yet publicly demonstrated it has solved.
What This Story Actually Teaches Founders
The instinct, watching a company as well-resourced and as talent-dense as OpenAI, is to assume its strategic moves are automatically sound simply because of the scale of capital and expertise behind them. The more useful reading treats OpenAI's hardware bet as an unusually well-documented, real-time case study in a decision every growing company eventually faces: what to do when your primary business model, however large, is not producing the kind of durable, expanding margin your cost structure actually requires.
The instructive part is not whether OpenAI succeeds. It is the reasoning trail visible in how the company has responded to that pressure: acquiring genuine hardware expertise rather than building it from zero internally, aiming at a differentiated economic model (owned hardware plus recurring services) rather than simply raising prices on the existing offering, and choosing to delay a high-profile launch by the better part of a year specifically to avoid repeating the well-documented failure modes of the two most recent competitors in the same category. Those are each reasonable, disciplined responses to a genuinely hard structural problem. Whether they add up to a trillion dollars of value, or to an expensive detour that further strains an already stretched balance sheet, is a question the market will not be able to answer with any confidence until a real product actually ships, currently expected no earlier than February 2027.
Five Things Worth Taking From This
A growing top line does not automatically fix a broken cost structure. OpenAI's revenue is expanding rapidly and its losses are expanding just as fast alongside it, because the core cost driver, inference compute, scales directly with the usage that also drives revenue. Growth alone does not resolve that kind of structural mismatch. A genuinely different economic model is sometimes required, not just more of the same model at greater scale.
Acquiring expertise can be faster and more credible than building it internally, but it does not remove execution risk. OpenAI bought genuine, proven hardware talent in Ive's team rather than assembling a hardware division from scratch. That is a sound instinct. It has not, on its own, resolved the underlying technical and product problems that two prior, well-resourced competitors in the same category also failed to solve.
A bold public number is not the same as a validated business case. Altman's trillion dollar framing for the io acquisition is a compelling narrative, and narratives matter for fundraising, talent recruitment, and market perception. It is not, by itself, evidence that the underlying unit economics of the product have been solved.
Delaying a launch to get it right is often the more disciplined choice, even when it looks like a setback publicly. OpenAI pushing its hardware release from 2026 to no earlier than February 2027, specifically to work through personality, privacy, and power issues, is arguably the correct response to watching two recent competitors fail on exactly those points. Public pressure to ship on the original timeline is rarely a good reason to ignore genuine, unresolved technical risk.
A single new product line is rarely a complete fix for a structural business model problem, and treating it as one is a real risk. Even in the most optimistic scenario for OpenAI's hardware ambitions, the core software business still needs to close a very large gap between revenue and cost on its own. A new revenue stream is most valuable as a genuine diversification of a fragile business, not as a singular rescue plan the entire company's survival is staked on.
OpenAI's hardware bet is, at this point, still mostly a bet. The prototypes exist. The trillion dollar framing exists. The shipping date, repeatedly, does not yet. What the story reliably offers right now is not a verdict on whether the device will work, but an unusually transparent look at what it actually looks like when one of the most well-funded companies in the world tries to build its way out of a business model that, on its current trajectory, does not yet add up.