#The AI Takeover Is Here: How 'Spud' Could Be OpenAI's Most Powerful Weapon Yet

TL;DR (Direct Answer): On March 24, 2026, OpenAI completed pre-training on a new model internally codenamed "Spud" — and Sam Altman told employees in an internal memo that it could arrive in "a few weeks" and "really accelerate the economy." The same week, OpenAI killed Sora — its AI video generator that generated $15 million in losses per day — winding down its Disney partnership and freeing up compute for Spud in one move. Altman simultaneously handed off direct oversight of safety and security teams so he could focus on "building datacenters at unprecedented scale." OpenAI's product organization, led by Fidji Simo, was renamed "AGI Deployment." The company confirmed plans for a desktop superapp that merges ChatGPT, its Codex coding agent, and the Atlas browser into one unified platform. OpenAI employees have hinted that Spud contains a capability that is "very different from what we've seen before." The backdrop: Anthropic is now capturing 73% of first-time enterprise AI spending. Claude overtook ChatGPT as the most downloaded app in the United States in March 2026. ChatGPT's traffic share has hit its lowest point since 2023. And OpenAI is projecting losses of $14 billion in 2026, with profitability not expected until 2029 at the earliest. Spud is not just a new model. It's OpenAI's answer to an existential competitive moment — and Altman says it's almost ready. This post breaks down everything: what Spud is, what it's replacing, what the superapp means, and why the timing of all of this should tell you something important.

#The Week OpenAI Decided to Stop Hedging

There is a version of the OpenAI story from the past eighteen months that reads as a company confidently expanding in every direction simultaneously. New video tools. New hardware partnerships. New browsers. New shopping features. New standalone apps for developers, creators, and consumers. A bet, as Altman himself described it, on "a series of startups" within one organization.

That version of the story ended on March 24, 2026.

In the span of four days, OpenAI killed its most high-profile consumer product, disbanded the CEO's direct oversight of safety, renamed its product organization after the goal it's now claiming to actively pursue, and told employees it had completed training on a model that can change the economy. Not described as "powerful" or "capable" or "state of the art." Described as capable of accelerating the economy. That's Altman using the language of civilization-scale impact for what is, from a product standpoint, the next version of ChatGPT.

"Things are moving faster than many of us expected," Altman wrote in his internal memo.

The week's announcements read as a company shedding weight before a sprint. Every resource being reallocated to Spud. Every product that wasn't generating revenue — gone. Every organizational layer that wasn't directly pointed at deployment — renamed, restructured, or handed to someone else. Whether Spud delivers on the framing surrounding it is a question only the next few weeks will answer. But the preparation tells you how seriously OpenAI's leadership believes in what they've built.

#What Is Spud, Actually?

#The Name Means Nothing. The Framing Means Everything.

Spud is OpenAI's internal codename for its next major model, which completed pre-training on approximately March 24, 2026, according to a report by The Information. The name is a placeholder — functional, unromantic, and almost certainly not the name it ships under. Whether this ends up being called GPT-5.5, GPT-6, or something else entirely is currently unknown. OpenAI has given no guidance on naming.

What we know about Spud from confirmed reporting and credible sources:

Spud is being trained at the Stargate facility in Abilene, Texas — the first Stargate site — using over 100,000 H100 GPUs. The parameter count has not been disclosed. Neither has whether it is a reasoning model, non-reasoning, or something with a new architecture. OpenAI employees have hinted that Spud contains a capability that is "very different from what we've seen before," though no specifics have been shared publicly. Altman has described it to staff as a "very strong model" that could "really accelerate the economy." A public release is expected within weeks — possibly late March or early April 2026.

The strategic function Spud is meant to serve is clearer than its technical specifications. OpenAI's planned desktop superapp — ChatGPT, Codex, and the Atlas browser merged into one unified platform — needs a model powerful enough to justify the architecture. An agentic environment where AI writes code, browses the web, orchestrates multi-step workflows, and operates autonomously on a user's computer requires a foundation model that can do all of those things reliably. Spud is expected to be that foundation.

#The Renaming That Isn't Just Semantic

OpenAI's product organization, led by Fidji Simo, was renamed "AGI Deployment" as part of the same week's announcements. This is worth dwelling on for a moment because corporate renamings are usually exactly as boring as they sound — except when they aren't.

Renaming a product organization "AGI Deployment" is not a naming exercise. It is a public internal declaration: OpenAI no longer considers itself in the phase of developing toward AGI. It considers itself in the phase of deploying it. That self-positioning is either the most audacious corporate communication of the year or a statement that will age very badly very quickly. Altman is betting on the former.

The context for evaluating that claim: ARC-AGI-3, which launched on March 25, 2026, drops AI agents into novel environments with no instructions and no win conditions, requiring genuine exploration and adaptive planning. Frontier AI models — including the best versions of GPT, Claude, and Gemini — scored under 1% on it. Humans scored 100%. On the most rigorous public benchmark for general intelligence, frontier AI models are nowhere close to human-level performance.

AGI Deployment is a name that describes where OpenAI wants to be, not necessarily where it is. Whether Spud changes that calculus is the most important question in AI right now.

#Why OpenAI Killed Sora — and What That Tells You About the Whole Strategy

#The Numbers That Made the Decision Inevitable

Sora launched in September 2025 with genuine fanfare. It briefly hit number one in the Apple App Store. It had a Disney deal that included Mickey Mouse licensing and a $1 billion stake in OpenAI. It generated breathless coverage about the future of AI video.

Then usage flatlined. Monthly downloads plummeted 32% in December 2025 — a month typically boosted by new smartphones and holiday app downloads. In January 2026, installations dropped another 45% to 1.2 million. Consumer revenue fell 32%. The platform's total lifetime revenue was $1.4 million from 9.6 million downloads — a coverage rate of less than one percent of actual operating costs.

Sora reportedly generated around $15 million in losses per day. Every GPU rendering video frames was unavailable for ChatGPT requests — and ChatGPT, with 900 million weekly active users and 50 million paying subscribers, generates the overwhelming majority of OpenAI's revenue. At a company projecting $14 billion in total losses for fiscal 2026 and expecting to need over $207 billion in additional funding by 2030, the decision to kill a product burning $15 million per day to free up compute for the model Sam Altman describes as economy-accelerating is not a difficult one. It's arithmetic.

OpenAI also wound down its Disney partnership with the Sora shutdown. Disney had previously agreed to license iconic characters for use on Sora and to take a $1 billion stake — entirely in stock warrants rather than cash. The collapse of that deal signals something the Sora revenue numbers confirm: video generation at consumer scale is not currently a viable AI business model. The compute required for high-quality video is simply too expensive relative to what users will pay.

OpenAI now reportedly plans to fold video generation into the main ChatGPT app — treating it as a feature rather than a product. That's the correct architecture lesson from Sora's failure: AI-generated video has a place, but its place is as one capability inside a larger productivity platform, not as a standalone consumer application that has to justify its own infrastructure costs.

#The Internal Diagnosis Nobody Was Saying Out Loud

The Sora shutdown was the most visible product casualty of a broader internal diagnosis that Fidji Simo put in writing in her March 16 memo to employees. "We realized we were spreading our efforts across too many apps and stacks," she wrote. "That fragmentation has been slowing us down and making it harder to hit the quality bar we want."

Simo described Anthropic's success to employees as a "wake-up call." Anthropic is now capturing 73% of all spending among companies buying AI tools for the first time, while OpenAI has fallen to around 27%. Claude overtook ChatGPT as the most downloaded app in the United States in March 2026. The competitor Altman spent years publicly pitying for its safety-focused caution is winning the market that matters most — enterprise and developer adoption — by doing the thing that is unglamorous and hard: staying focused on text and code while OpenAI was busy filming Mickey Mouse.

"We cannot miss this moment because we are distracted by side quests," Simo told employees. That phrase — side quests — is the internal language for everything that wasn't ChatGPT, Codex, and the core reasoning model. Shopping features. Video generation. Hardware devices. Browser-as-standalone-product. All of it, in retrospect, a distraction from the thing Anthropic never stopped doing.

#The Superapp That Has to Work

#What It Actually Is

OpenAI is building a desktop superapp that will merge ChatGPT, the Codex coding platform, and the Atlas browser into a single unified application. Greg Brockman is temporarily leading the technical overhaul. Fidji Simo is handling distribution. The mobile ChatGPT app remains unchanged.

The architecture is worth understanding in detail because it describes what OpenAI is trying to become. ChatGPT is the orchestration layer — the natural language interface where users describe what they want to accomplish. Codex is the execution layer for software development — capable of writing new features, fixing bugs, and completing complex engineering tasks end-to-end. Atlas is the browsing and research layer — a Chromium-based browser with a memory system that stores context across sessions, and an agent mode that can complete tasks within web pages autonomously.

Put them together and you have something genuinely different from any current AI product: an environment where a user can say "organize my project, update the documentation, research the three competitors we discussed last week, and push the changes to GitHub" — and the platform knows the project context, writes and tests the code, browses to gather competitive intelligence, and executes the commit. Without switching applications. Without losing context. Without manually coordinating between tools.

That's the theoretical promise. The execution challenge is significant. OpenAI has already launched several standalone products over the past year that didn't resonate — Sora being the most prominent but not the only example. Agent mode, launched as essentially a browser-use feature, reportedly lost most of its users because nobody understood what it was actually for. Building an integrated platform is architecturally harder than building individual applications, because every component has to work together under conditions that no single application's testing ever encounters.

#The Competitive Pressure That Made It Urgent

The superapp strategy is explicitly a response to Anthropic's focus. Anthropic has placed far fewer product bets than OpenAI, consistently zeroing in on the enterprise and coding market — deliberately staying away from audio, image, and video generation while OpenAI was pursuing all three.

The results are now visible in the market share numbers. Claude Code — Anthropic's agentic coding tool — became the reference implementation for AI-assisted software development that enterprise teams trust and use in production. Cowork — Anthropic's autonomous computer agent — addressed the agentic desktop use case that OpenAI's various agent features were trying to crack without a coherent platform underneath them.

OpenAI's superapp is the structural answer to that focused competition: instead of matching Anthropic product-for-product, consolidate everything into one platform that's more capable, more integrated, and harder to compete with than any individual tool. Whether the integration quality is high enough to win back the enterprise and developer adoption Anthropic has taken is the question that will define OpenAI's 2026 competitive position.

#The Safety Handoff Nobody Should Ignore

Buried in the same week's announcements — overshadowed by Sora's death and Spud's birth — was a structural change at OpenAI that deserves more attention than it's getting.

Sam Altman relinquished direct oversight of OpenAI's safety and security teams. His stated reason: to focus on raising capital, supply chains, and "building datacenters at unprecedented scale." Safety oversight has been delegated to others in the organization.

Read that in isolation and it sounds like routine organizational delegation. Read it in context and the timing is jarring. OpenAI is renaming its product organization "AGI Deployment." It is completing training on a model employees describe as containing capabilities "very different from what we've seen before." It is weeks away from releasing that model publicly. And at precisely that moment, the CEO — who has previously described AI safety as an existential priority — is stepping back from direct safety oversight to focus on data centers.

In February 2026, OpenAI dissolved its "Mission Alignment" team, which had been responsible for public communications about the company's safety commitments. OpenAI is also actively fighting the Pentagon's demands to remove safety guardrails from Claude — a battle where Altman publicly backed Anthropic's position while simultaneously signing his own Pentagon deal with different contractual framing.

The honest question the timing raises: is the safety handoff a routine organizational evolution, or is it the CEO creating decision-making space at the moment the company is about to ship something significant? The answer will become visible in what OpenAI's external safety communications look like in the months following Spud's release.

#The IPO Shadow Over All of It

One more piece of context that colors everything above: OpenAI is preparing for a potential IPO in 2026. The company is valued at approximately $730 billion following a $110 billion funding round closed in February 2026. HSBC analysts estimate it may need over $207 billion in additional funding by 2030 — and it won't reach profitability until 2029 at the earliest under current projections.

An IPO requires telling a coherent story about what the company is and where it's going. A scattered portfolio — video generators losing $15M a day, shopping features nobody used, agent modes without clear use cases — does not tell a coherent story. A unified superapp powered by a next-generation model that the CEO says can accelerate the economy, built on a $553 billion Oracle infrastructure partnership, in active deployment with 900 million weekly active users — that story is much cleaner.

The Sora shutdown, the superapp announcement, the AGI Deployment renaming, the safety handoff, and the Spud pre-training completion all happened in the same week. That's not coincidence. That's a company organizing itself for a very specific public moment, and doing it quickly enough that competitors don't have time to respond before Spud lands.

#The State of Play Going Into Spud's Launch

#What This Means Right Now

If you use ChatGPT: Spud is the next model you'll be using — probably within weeks. Based on what OpenAI employees have suggested and what the Stargate training infrastructure implies, this is likely the most capable model OpenAI has ever released. The practical experience of using it inside the superapp architecture — where ChatGPT orchestrates Codex and Atlas in a single environment — will be meaningfully different from anything currently available.

If you build on the OpenAI API: The superapp consolidation will affect how OpenAI exposes its models and capabilities. The shift toward an "agentic" architecture — where the platform autonomously coordinates tools rather than responding to individual prompts — signals that API access patterns will evolve. Watch the Codex API specifically, as it's the component most directly aimed at developer workflows.

If you're currently using Claude or Anthropic products: The competitive dynamic is about to get more interesting. Anthropic has had a clear runway over the past six months while OpenAI was managing product fragmentation. Spud's release compresses that runway significantly. The 73% enterprise spending share Anthropic currently holds is real and hard-won — but it was won against an OpenAI that was, by its own application chief's admission, distracted by side quests. The version of OpenAI that ships Spud and the superapp is more focused and more dangerous as a competitor.

If you're watching for signs of real AGI progress: The ARC-AGI-3 benchmark result — frontier models at 0.37%, humans at 100% — is the honest check on every claim about imminent transformative AI. Renaming your product organization "AGI Deployment" does not constitute AGI. The gap between what Spud will be able to do and what a human can do across genuinely novel environments remains enormous. Track what the benchmark scores look like after Spud releases. That's the signal, not the name.

#FAQ

What is OpenAI Spud and when is it releasing?
Spud is the internal codename for OpenAI's next major model, which completed pre-training on approximately March 24, 2026. Sam Altman told employees it would be available in "a few weeks" — suggesting a late March or April 2026 release. It will likely be called GPT-5.5 or GPT-6, though OpenAI hasn't confirmed the name. Employees have described it as containing capabilities "very different from what we've seen before," and Altman has said it can "really accelerate the economy."

Why did OpenAI shut down Sora?
Sora was losing approximately $15 million per day in operating costs while generating $1.4 million in total lifetime revenue from 9.6 million downloads — a coverage rate under 1% of actual costs. Monthly downloads had fallen 45% between November 2025 and January 2026. Every GPU dedicated to video rendering was unavailable for ChatGPT requests, which generate the overwhelming majority of OpenAI's revenue. Shutting down Sora freed compute directly for Spud. OpenAI plans to fold video generation into the main ChatGPT app as a feature rather than a standalone product.

What is OpenAI's superapp?
The superapp is a unified desktop application that will merge ChatGPT, OpenAI's Codex coding agent, and the Atlas browser into a single platform. ChatGPT serves as the conversational orchestration layer, Codex handles software development and automation, and Atlas handles web browsing and research with a persistent memory system. The result is designed to be an end-to-end agentic environment where AI can complete complex multi-step tasks — writing code, browsing the web, analyzing data — without the user switching between applications. The mobile ChatGPT app remains unchanged.

Why did Altman hand off safety oversight?
Altman stated the reason as a need to focus on "building datacenters at unprecedented scale," capital raising, and supply chains. He has delegated safety and security team oversight to others at OpenAI. The timing — immediately before Spud's release, after the Mission Alignment team was disbanded in February, and while the product org was renamed "AGI Deployment" — has raised questions in the security and AI safety community about whether the structural change reduces rigor at a critical moment.

What does "AGI Deployment" mean as a product org name?
It signals that OpenAI's internal framing has shifted from building toward AGI to actively deploying it. Whether that framing reflects technical reality is contested — ARC-AGI-3 shows frontier AI models at under 1% performance on tasks requiring genuine novel problem-solving, compared to human 100%. The rename is best understood as an expression of where OpenAI wants to position itself for its IPO and for Spud's release, rather than as a verified claim about what Spud can do.

Why is Anthropic winning the enterprise market right now?
Anthropic has captured approximately 73% of first-time enterprise AI spending in 2026, compared to OpenAI's 27%, primarily through the success of Claude Code and Cowork — products that stayed focused on developer productivity and enterprise workflows while OpenAI was launching Sora, shopping features, and standalone browsers. Anthropic's narrower product surface area meant higher quality and trust in the specific use cases enterprises care about. Claude overtook ChatGPT as the most downloaded app in the U.S. in March 2026. OpenAI's superapp and Spud launch are its structural response to that competitive gap.