OpenAI Is Not a Lab Anymore. It's a Media Company.

The easiest way to misread OpenAI right now is to keep evaluating it as an AI laboratory that builds impressive models. That framing is three strategic pivots out of date. The more useful frame — the one that explains the content licensing agreements, the advertising launch, the social video app, the voice mode rollout, the $250 million News Corp deal, and the acquisition of a tech talk show — is that OpenAI has become a media company. Not metaphorically. Structurally. And the industry it is reshaping most urgently is not enterprise software. It is publishing.

The AI community is treating the TBPN acquisition as a quirk. Publishers should be treating it as a confirmation.

What the TBPN Deal Actually Signals

On April 2, 2026, OpenAI acquired TBPN — the Technology Business Programming Network — a daily live tech talk show hosted by John Coogan and Jordi Hays that had built a devoted Silicon Valley audience through unfiltered, insider-flavored coverage of the technology industry. The acquisition price was reported in the low hundreds of millions of dollars for a startup that generated approximately $5 million in advertising revenue in 2025 and was on track for $30 million in 2026. That multiple — roughly 40 to 60 times 2025 revenue — is not how you price a communications tool. It is how you price distribution.

TBPN had hosted Mark Zuckerberg, Satya Nadella, Marc Benioff, and Altman himself, and its value was exactly that access — the credibility of a show where powerful people spoke candidly to fellow insiders rather than to PR-vetted press. That is an editorial asset. OpenAI’s CEO of Applications, Fidji Simo, described TBPN as bringing “strong editorial instincts, deep audience understanding, and a proven ability to convene influential voices across tech, business, and culture.” That is a media company description of a media asset. Simo also noted, with striking candor, that “the standard communications playbook just doesn’t apply to us” — an acknowledgment that OpenAI has outgrown the traditional technology company communications posture and needs something closer to editorial infrastructure.

The show will report to Chris Lehane, OpenAI’s chief global affairs officer. TechCrunch noted that Lehane is the political operative who invented the phrase “vast right-wing conspiracy” as a press management tool for the Clinton White House, and who later ran Fairshake, the crypto industry super PAC that spent hundreds of millions of dollars to shape the 2024 election. Placing an editorially independent tech talk show under a political operative who specializes in narrative management is not a communications decision. It is an information environment decision. Margaret O’Mara, a University of Washington professor who studies the history of technology and politics, told NPR it fits the mold of tech companies launching vehicles that “advance the goals of their owners and sponsors” — while noting the risk that TBPN could appear too controlled despite assurances of independence.

That tension is real and will deepen. But the more important observation is structural. OpenAI has now made its first acquisition of a media company. It will not be the last.

The Anatomy of a Media Pivot

The TBPN acquisition is the clearest signal, but it did not come from nowhere. OpenAI’s trajectory since 2024 follows a recognizable pattern — the same pattern that defined the early-stage development of every major media platform of the internet era. First, acquire or license the content that creates value for users. Then build the distribution infrastructure that locks those users in. Then monetize the audience at scale.

OpenAI has been executing all three phases simultaneously, and each deserves a clear-eyed look.

On content acquisition: News Corp signed a five-year licensing deal with OpenAI worth over $250 million — the Wall Street Journal, Barron’s, the New York Post, and News Corp’s other properties now feed into OpenAI’s training and product pipelines. Robert Thomson, News Corp’s CEO, called it a deal that would “set new standards for veracity, virtue and value in the digital age.” The Atlantic and Vox Media signed separate content and product partnerships with OpenAI on the same day in May 2024, giving OpenAI access to their archives and allowing their content to surface within ChatGPT. Condé Nast followed with a multi-year licensing arrangement. The Guardian entered a partnership. So did the Financial Times, Fortune, Time, and dozens of others. The speed and breadth of these agreements — executed across a single calendar year — reflects an organization building a content library with the urgency of a platform preparing for launch, not a laboratory seeking training data.

On distribution infrastructure: ChatGPT has reached 900 million weekly active users as of early 2026, making it the most downloaded application of 2025, surpassing both TikTok and Instagram. It processes approximately 2.5 billion daily prompts. That is not a tool. That is an audience. An audience that, according to a large-scale OpenAI and Harvard analysis of user conversations, spends 49% of its time asking for information and guidance — the exact behavioral profile that makes a platform valuable to advertisers and content partners alike. ChatGPT now accounts for 60.4% of AI search market share and, combined with Microsoft Copilot which runs on the same underlying model, controls approximately 73% of the emerging AI search market.

On monetization: OpenAI launched advertising in ChatGPT for free and low-tier users in January 2026. The ads pilot crossed $100 million in annualized revenue within its first six weeks. OpenAI has told investors it projects ad revenue of $2.5 billion in 2026, $11 billion in 2027, and $53 billion by 2029 — projections that assume the platform reaches 2.75 billion weekly users by 2030. Evercore ISI analyst Mark Mahaney projects OpenAI’s advertising business could reach $25 billion annually by 2030, placing it in the same tier as established advertising giants. Sam Altman once said he hated ads as “an aesthetic choice.” That aesthetic has been overridden by the economics of running a platform at this scale — a transition that every major media company in the internet era has made in exactly this sequence.

Sora Was the Experiment That Told You Everything

The story of Sora is worth telling carefully because it is frequently misread as a failure. It is more useful to read it as a proof of concept that ran into the wrong economics.

OpenAI launched Sora’s standalone app in September 2025 — a social video platform where users could create AI-generated videos and share them in streams that resembled TikTok in design. Multiple outlets including Wired noted the explicit stylistic similarity to TikTok. The app reached the top of the iOS App Store’s photo and video category within a single day of launch. It attracted Android, had a Sora 2 model update in September 2025, and was progressing toward a landmark content deal. In December 2025, Disney announced a $1 billion equity investment in OpenAI that included a licensing agreement to bring Disney characters into Sora — with curated Sora-generated videos slated to appear on Disney+ in early 2026.

A laboratory does not sign a $1 billion content deal with Disney to put Pixar and Marvel characters in a social video app. A media company does.

OpenAI shut Sora down on March 24, 2026, with the web and app experiences set to discontinue on April 26, 2026. Bill Peebles, OpenAI’s head of Sora, acknowledged publicly that “the economics are completely unsustainable.” The video inference costs — the compute required to generate, serve, and distribute AI video at social-media scale — could not be bridged by in-app credit purchases at current compute prices. Disney exited the deal. The experiment ended.

But the experiment’s ending should not obscure what the experiment revealed about intention. OpenAI was not building a video generation feature for ChatGPT. It was building a social video distribution platform with content licensing from one of the most powerful intellectual property holders on earth. That is not a model capability. It is a media architecture. The economics failed it. The media ambition was clear.

The Content It Licenses Is the Content It Displaces

The most uncomfortable dynamic in OpenAI’s media pivot is that it is happening on both sides of the publishing ledger simultaneously. OpenAI is paying publishers for content while building a platform that redirects the audience that content depends on. Those two activities are not in tension — they are structurally connected. The licensing deals are the input. The platform is the output. The publisher’s content produces value inside ChatGPT, and the publisher’s audience finds less reason to visit the publisher’s own property.

The data on this displacement is no longer speculative. Chartbeat’s analysis covering December 2024 through December 2025 found that referral traffic from traditional search engines declined 60% for small publishers over two years. Organic click-through rates on queries featuring Google AI Overviews dropped from 1.76% in June 2024 to 0.61% by September 2025. AdExchanger reported that some well-known tech-focused publications have seen search referrals drop by as much as 85%, with smaller publishers being forced to shut down and “more will no doubt join them in 2026.”

ChatGPT did send 1.2 billion outgoing referrals to publisher sites between September and November 2025 — a 52% year-over-year increase — and that number matters. But AI platforms as a category still account for only 1% of overall web traffic across major industries, according to Conductor. Despite 200% growth in ChatGPT referrals throughout 2025, chatbots still account for less than 1% of all pageview referrals to publishers, while the organic traffic they are replacing was doing meaningful, monetizable work. The replacement is not proportional. The extraction is happening faster than the compensation.

Semrush’s analysis of 17 months of clickstream data found that ChatGPT’s search feature is active on only 34.5% of queries as of February 2026 — down from 46% in late 2024 — meaning the majority of responses still draw entirely on training data rather than live web retrieval. That means most ChatGPT answers are delivered from a library built on publisher content, without a click leaving the platform, without attribution generating a visit, and without a view generating ad revenue for the organization that created the underlying knowledge. The licensing deals pay a fraction of publishers for a fraction of that content. The vast majority of publishing that feeds ChatGPT’s knowledge base is simply consumed.

Voice Is the Quietest Part of the Pivot

The advertising and acquisition stories have received attention. The voice mode story has not, and it is arguably more structurally significant for publishers in the medium term.

OpenAI rolled out Advanced Voice Mode — a native, end-to-end multimodal voice system built on GPT-4o — to ChatGPT Plus users beginning in 2024, expanding to all free users in 2025 via a GPT-4o mini version. The system operates at human-conversation speed, with average response latencies around 320 milliseconds. It detects emotional tone, handles interruptions, manages pauses, and produces responses that are qualitatively closer to a radio host or a podcast than to a search result. OpenAI has continued releasing audio model updates, including specialized models for custom voice agents, transcription, and real-time interaction — building an audio infrastructure that resembles, in its ambition, the early API layers of what eventually became podcast distribution platforms.

The voice medium matters to publishers because it is the one format where the source of information is most invisible to the user. When you read a news article, the masthead tells you who wrote it. When a voice interface answers your question in natural spoken language, there is no masthead. The content has been absorbed into the interface. Whatever attribution model publishers negotiated in their text-based licensing deals does not transfer to an audio response. The voice mode is where licensed content most completely disappears into the platform — and where the platform most completely becomes the publication.

The Media Company Playbook, Accelerated

The sequence OpenAI has executed is not novel. It is the standard platform playbook, running at compressed speed with better capital and more favorable regulatory conditions than any predecessor operated under.

Facebook built its media identity by aggregating publisher content, directing audience to it long enough to build user habits and advertising scale, and then progressively restricting organic reach to monetize the distribution it had captured. Google did the same through search, and is currently executing a second version of it through AI Overviews. Twitter under Jack Dorsey positioned itself as real-time news infrastructure and monetized the journalists and publications that depended on it for distribution. Each of these companies told publishers, at the beginning of the relationship, that the platform was a partner. Each eventually revealed that the publisher’s content was the product, and the publisher’s audience was the customer.

OpenAI is running this playbook at what looks like a five-year compression. The content licensing phase, which took Facebook nearly a decade to navigate, was completed in approximately 18 months. The advertising infrastructure, which Twitter never adequately built, launched at $2.5 billion in projected year-one revenue. The owned media asset — TBPN — was acquired before most platforms of comparable audience size had even considered editorial ownership.

The difference is that OpenAI has done all of this in public, with explicit explanations, with willing publisher partners, and with a stated mission to “benefit all of humanity.” That mission statement is doing substantial work in preventing the media industry from applying the analytical frameworks it learned, painfully and expensively, from the previous generation of platform companies. Publishers that signed content licensing deals with OpenAI in 2024 should be asking, in 2026, whether the terms of those deals account for a platform that is now also running advertisements, building owned media, and projecting $53 billion in ad revenue by 2029.

What Publishers Should Actually Do With This

None of this means OpenAI is acting in bad faith. It means OpenAI is acting in its own interest, on a timeline that is moving faster than most of its content partners anticipated when they signed their agreements. Understanding that interest clearly is the precondition for any strategic response.

The first thing publishers need to accept is that the licensing deal economics they negotiated in 2024 were set at a moment when OpenAI’s platform identity was genuinely uncertain. News Corp’s $250 million, five-year deal with a company that has now crossed $25 billion in annualized revenue and is projecting $53 billion in ad revenue from the platform that runs on News Corp’s content is a different deal than it looked like when it was signed. Renegotiation should be treated as a structural necessity, not a negotiating preference.

The second thing is that the referral traffic replacement argument — the idea that ChatGPT will eventually send enough clicks to offset what it takes from search — requires scrutiny. ChatGPT referred traffic to US retail sites converted at 11.4% compared to 5.3% for organic search, which is a meaningful quality signal. But quality cannot compensate for volume when volume is declining at 60% for small publishers and the referral alternative accounts for less than 1% of overall traffic. The gap is not narrowing fast enough to rely on the replacement narrative as a strategy.

The third thing — and the one the industry has been most reluctant to name clearly — is that the question of whether OpenAI is a lab, a platform, or a media company is not an identity question. It is a regulatory and competitive question. Labs get different treatment than platforms. Platforms get different treatment than media companies. The frameworks that govern content liability, advertising standards, copyright, and public interest obligations differ substantially across those categories. OpenAI’s current hybrid identity allows it to invoke lab status when convenient, platform status when useful, and media company credibility when it needs the TBPN acquisition to make sense. That flexibility has real value — and it has been underpriced by every public interest institution that should be pricing it.

The Lab Framing Was Always a Choice

OpenAI was founded as a nonprofit research organization in 2015. That origin story has remarkable staying power in public discourse, surfacing constantly in coverage that still frames the company’s decisions as departures from a research identity, rather than as expressions of a commercial one that has been visible for years.

OpenAI’s annualized revenue crossed $25 billion by February 2026, up from $20 billion at the end of 2025. It has raised $122 billion in its most recent funding round at an $852 billion post-money valuation. It has 92% of Fortune 500 companies using its products, 9 million paying business users, and a consumer product that is the most downloaded application on the planet. It is preparing an IPO. It has launched advertising. It has acquired a media company. It is projecting ad revenue that would place it alongside established advertising giants within four years.

None of that is a lab. Some of it was never going to be a lab. The nonprofit framing served a fundraising and talent acquisition purpose in 2015 and a regulatory positioning purpose for years afterward. The lab identity is now a legacy brand asset — the kind of reputational capital that gets spent in regulatory hearings and op-eds while the organizational reality has moved on.

The media company identity is not a rebranding. It is an observation about what OpenAI has actually built. A platform with 900 million users, a social video ambition, an advertising business, an owned editorial property, a voice distribution layer, and content licensing agreements with the majority of the English-language premium publishing world is a media company by any operational definition — regardless of what its 2015 charter says about the beneficial development of artificial general intelligence.

Publishers have spent the last two years wondering whether to partner with OpenAI, license to it, sue it, or ignore it. The more useful question, the one the industry has been reluctant to ask with the full weight it deserves, is simpler: what do you do when your biggest content partner is also your most direct structural competitor? That question does not have a comfortable answer. It does have an urgent one.