OpenAI rides the AI wave into major expansion: video generation, checkout shopping, social networking, and hardware devices—everything is on the table. However, after Sora burned through an astronomical amount and ended in a quiet fade-out, a $1 billion Disney partnership fell apart, the $500 billion Stargate plan stalled, and Nvidia’s investment commitment shrank significantly, Forbes took a full inventory of OpenAI’s products and deals that it had “promised but didn’t deliver.”
(Background: OpenAI reveals that the “AI bubble is bursting”: Sora is halted, Disney withdraws $1 billion, a Pentagon controversy, a $11.5 billion loss in a single quarter)
(Additional context: Oracle may have delayed the construction of OpenAI’s data center; AI concept stocks all plunged, and after bitcoin’s sudden drop it rebounded to $90,000.)
Table of contents
Toggle
- – Failed product section
- The video dreams burned by Sora + Disney
- NSFW mode: controversy shelved indefinitely
- Real-time checkout shopping: conversion rate only 1/3 of e-commerce platforms
- GPT-4o: stripped by its own successor
- – Stalled deals section
- Stargate: the $50 billion plan reduced to a pipe dream
- Nvidia: $100B investment intent shrinks to $30B
- AMD: a chip deal worth 6 billion watts—so far, zero shares received
- AI bubble narrative heats up
Fidji Simo, CEO of OpenAI’s Applications division, defended the company’s product strategy earlier this January: ads, shopping, health, social networking, browsers, physical devices, video generation, the App Store marketplace—everything they wanted to do.
However, just two months later, OpenAI made a sharp turn: it shut down Sora and canceled the Disney partnership. Under pressure from a potential IPO, the company is shifting from “doing everything” to financial discipline, concentrating resources on coding and enterprise productivity tools. Even though OpenAI reached $13 billion in revenue in 2025, it remains deep in losses, while competitor Anthropic is rapidly catching up.
Forbes compiled a list of OpenAI’s “product graveyard”—the plans and deals that were once announced with great fanfare, but still haven’t been delivered.
– Failed product section
The video dreams burned by Sora + Disney
Last December, OpenAI and Disney reached a “milestone” partnership: Disney invested $1 billion and licensed 200 characters for Sora to use. Disney CEO Bob Iger was on board. But this year in March, everything abruptly ended.
At its peak, Sora burned $15 million per day; however, according to data from Sensor Tower and Appfigures, the total in-app purchase revenue over its entire lifecycle was actually less than $3 million. A Disney spokesperson said: “We respect OpenAI’s decision to exit the video generation business.”
Further reading: OpenAI announces it is shutting down the Sora app; Disney’s $1B partnership deal falls apart: a major positioning failure for social platforms
NSFW mode: controversy shelved indefinitely
Last October, Altman proposed allowing OpenAI’s ChatGPT to host adult-related conversations for verified adult users. The move immediately triggered strong backlash from employees and investors. According to a March report by The Financial Times, the feature has been “shelved indefinitely.”
There are also technical challenges in filtering illegal content. OpenAI said it hopes to conduct long-term research into the relevant impacts first.
Real-time checkout shopping: conversion rate only 1/3 of e-commerce platforms
Last October, OpenAI partnered with Walmart to list 200,000 products in ChatGPT, highlighting a “real-time checkout” feature. Retailers such as Shopify and Etsy also participated, but to a limited extent. Yet the data is unforgiving: the conversion rate for shopping through ChatGPT is only one-third of going directly to Walmart.com.
In March, OpenAI announced it was cutting this feature. Consumers now instead use each retailer’s dedicated app inside ChatGPT. OpenAI admitted: “The initial version didn’t deliver the flexibility we were looking for.”
GPT-4o: stripped by its own successor
In February, GPT-4o officially went offline. The model is known for a warm, upbeat personality, but it has also been criticized for “over-the-top flattery.” When it was first taken down in August last year, it sparked a user uprising, and OpenAI eventually brought it back.
A Reddit user’s comment summarized users’ feelings precisely: “Give 4o back to me. GPT-5 is wearing the skin of my dead friend.”
– Stalled deals section
Stargate: the $50 billion plan reduced to a pipe dream
Not long after Donald Trump’s second term began, OpenAI announced the $500 billion “Stargate” plan with Oracle and SoftBank. A year later: it’s stalled—no enough talent was brought in, no facilities were built, and the three partners were deeply divided.
OpenAI once tried to take over the construction itself, but later shelved the idea. The plan to expand a thousand-acre campus in Abilene, Texas, handled by Crusoe, was also canceled after the financing fell apart. Oracle claimed it had completed the leasing of an additional 4.5GW of capacity, but the overall outlook for the plan remains unclear.
Nvidia: $100B investment intent shrinks to $30B
Last September, Nvidia announced it was “interested in investing” $100 billion in OpenAI, without setting a timeline. Now, if the IPO goes through, this investment could be cut to just $30 billion. Nvidia’s annual report states “no guarantee” that the transaction will be completed.
This $30 billion is essentially a GPU procurement subsidy: enough to build about 1GW of data center capacity. The model mirrors how Nvidia invests in CoreWeave, Nebius, and Nscale: investing in its own customers, ensuring they have money to buy more GPUs.
AMD: a chip deal worth 6 billion watts—so far, zero shares received
In October last year, OpenAI and AMD reached an agreement: 160 million shares of AMD stock (at the time worth about $30 billion, now about 10% of AMD’s total share capital) in exchange for a 6GW AMD chip data center. But the terms were extremely harsh: the shares would only be paid out after 1GW went online, AMD’s stock price hit specific targets, and other vague milestones were met.
As of the end of 2025, the number of shares cashed out was: zero. AMD’s chip deliveries are expected to begin only in the second half of 2026. If the deal ultimately closes, AMD expects it will bring “tens of billions of dollars” in revenue.
AI bubble narrative heats up
OpenAI’s list of graveyard projects is not just one company’s story—it reflects a turning point across the entire AI industry, moving from “unlimited expansion” to “pragmatic contraction.” When even the flagship players with valuations above $1 billion have to cut back on money-burning businesses and tighten their fronts, market confidence in the AI narrative is being put through stress tests.
