The AI collaboration chain involving NVIDIA, OpenAI, and Microsoft is showing signs of strain due to diverging views on "execution capability" and "operational discipline."
On Wednesday, Bloomberg technology columnist Parmy Olson suggested in a recent opinion piece that Jensen Huang's hesitation regarding OpenAI is not merely a shift in sentiment for a single investment, but rather serves as a reminder to Microsoft that exclusive access to models and intellectual property does not automatically translate into product superiority; market competition is shifting from "which model is more powerful" to "which company can better implement it."
NVIDIA CEO Jensen Huang recently indicated to industry insiders that his previously announced $100 billion investment in OpenAI is "not binding," and he has privately criticized OpenAI for lacking commercial discipline. Huang denied being dissatisfied with OpenAI and told reporters last Saturday that "we will invest a lot of money," but he simultaneously chose to leave room regarding the commitment.
Parmy Olson believes this investment, originally tied to infrastructure expansion, is now more likely to materialize as a "multi-billion dollar" commitment during OpenAI's current funding round, potentially occurring before a potential IPO. The report also stated that OpenAI is discussing approximately $100 billion in financing with NVIDIA, Microsoft, and Amazon, which is separate from the previously proposed infrastructure transaction.
The more direct market impact falls on Microsoft. Despite Microsoft securing exclusive access to OpenAI's intellectual property and models until 2032 through a restructuring deal announced last October, and planning to integrate the latest GPT-5 model into Copilot by August 2025, user feedback on Copilot's product experience remains critical, and its feature set lags behind competitors, highlighting the contradiction of "holding top-tier models yet struggling to capitalize on the advantage."
Jensen Huang's "buyer's remorse," in Parmy Olson's view, is particularly noteworthy because the AI industry has grown "numb" to massive bets, making Huang's caution with commitments stand out. The Wall Street Journal noted that Huang not only emphasized the non-binding nature of the investment but also privately questioned OpenAI's operational and execution discipline, introducing variables into OpenAI's funding narrative and partner confidence.
For investors, this stance signals two things: first, capital and computing power planning surrounding OpenAI may still be undergoing rapid adjustments; second, even one of the core industry partners is reassessing risk exposure and return pathways.
Parmy Olson pointed out that Sam Altman's management style continues to cause external concern, including his dramatic ouster and subsequent return at the end of 2023, along with a series of complex, staggeringly large transaction arrangements. She wrote that these arrangements have committed OpenAI to computing resources worth up to $1.4 trillion, a scale approximately 100 times OpenAI's projected 2025 revenue.
On the product side, OpenAI's progress also appears rushed. Parmy Olson cited examples, noting that OpenAI attempted to build a developer marketplace through GPT Store and custom GPTs, but it "fizzled" due to a lack of clear strategy. Between "sustained leading model capability" and "volatility in commercialization and organizational execution," OpenAI's partners must price in the uncertainty.
Microsoft's capital returns on paper are exceptionally substantial. Parmy Olson wrote that Microsoft's early $13 billion investment in OpenAI now corresponds to about 27% equity, valued at approximately $135 billion, over ten times the original outlay; meanwhile, the restructuring agreement grants Microsoft a "cleaner" exclusive IP and model access arrangement until 2032.
However, at the product level, Microsoft has not fully capitalized on this advantage. Parmy Olson identified the key issue: since OpenAI continues to release what are arguably the world's most powerful AI models, why does Microsoft's flagship product Copilot, built on OpenAI's technology, still lag behind competitors? She wrote that user feedback on Copilot centers on it being "confusing, limited, and difficult to use," highlighting a disconnect between model quality and product implementation.
Competitive pressure is intensifying at the application level. Parmy Olson wrote that last month, Anthropic launched Claude Cowork, an application reportedly built using its own AI coding tools within 10 days, which can operate a personal computer, organize files, generate PowerPoint and Excel files from documents, and respond to LinkedIn messages after receiving authorization.
In contrast, she pointed out that despite Microsoft owning Windows, Office, and LinkedIn, Copilot cannot perform these capabilities. This gap makes it easier for the market to attribute the problem to Microsoft's internal R&D and productization mechanisms, rather than solely to the model itself.
Parmy Olson cited David Rainville, head of Sycomore Sustainable Tech, who stated that some industry observers are watching the leadership trajectory of Microsoft's AI projects. He indicated that if Microsoft fails to release a product equivalent to Claude Cowork within the next six months, "heads will roll," and noted a clear disconnect between Microsoft's "model quality" and "execution capability."
Comments