When Nvidia CEO Jensen Huang announced in September 2025 that his company would invest up to $100 billion in OpenAI, the tech world celebrated what appeared to be a historic vote of confidence in artificial intelligence. Six months later, the story reads rather differently. At the Morgan Stanley Technology, Media & Telecom Conference this week, Huang delivered a more measured message: the company will invest $30 billion in OpenAI's current funding round, but the $100 billion infrastructure plan is, in his words, "probably not in the cards."
This retreat deserves serious attention, not because it represents a loss of faith in AI but because it reveals something important about how disciplined capital allocation works in practice. The difference between what was promised and what will be delivered points to a market correction overdue in a sector that has spent the past two years burning through investor enthusiasm with sometimes reckless abandon.
The original arrangement, signed as a non-binding letter of intent, tied Nvidia's investment to OpenAI's deployment of new supercomputing infrastructure. The two companies promised to build 10 gigawatts of computing capacity over time, with Nvidia progressively investing as each phase came online. It was, in theory, a tidy arrangement that aligned incentives and demonstrated commitment. In practice, it never materialised. The agreement stalled. Internal doubts emerged. By November, Nvidia's regulatory filings were already signalling caution. This week's announcement simply formalised what the market had quietly suspected.
Huang's reasoning for the shift is worth examining closely. He points to OpenAI's planned IPO by late 2026 as the key reason the massive infrastructure deal no longer makes sense. Once OpenAI goes public, the logic goes, future investments can happen at market prices rather than at the negotiated rates of a private company. This is defensible, though it also reads as a convenient justification for stepping back from what some within Nvidia apparently saw as an overly generous arrangement.
The investment landscape has shifted fundamentally since September. Investor scrutiny of AI valuations has intensified. Questions about "circular funding"—where semiconductor makers invest in AI companies that then purchase their chips—have become difficult to ignore. Wall Street analysts, regulators, and prudent investors have all raised concerns that some of these intertwined relationships artificially inflate demand and obscure the true viability of AI businesses. Whether those concerns are entirely fair or somewhat overdrawn, Huang is plainly aware that Nvidia's credibility depends on appearing above such criticism.
Tellingly, Huang insists that Nvidia will remain deeply involved in OpenAI's growth. The $30 billion investment is substantial; he has described it as potentially the largest investment Nvidia has ever made in a single company. The company is also expanding OpenAI's access to computing capacity across Amazon Web Services, Microsoft Azure, and Oracle Cloud. This is not abandonment. It is recalibration. Nvidia has chosen to participate in OpenAI's success through equity rather than through an open-ended infrastructure commitment that kept the two companies entangled in ways that invited legitimate questions about conflicts of interest.
What makes this story more layered is that Huang is simultaneously doubling down on a different kind of infrastructure bet. Nvidia announced $4 billion in investments in photonics firms Lumentum and Coherent, companies developing optical and laser technologies essential to next-generation AI data centres. These deals include multibillion-dollar purchase commitments from Nvidia and future access rights to critical components.
Here lies a pragmatic argument for Huang's approach. The bottlenecks in AI infrastructure are becoming increasingly sophisticated. The constraint is no longer simply compute power; it is the optical networking, power delivery, and interconnect technologies that allow vast numbers of AI chips to communicate efficiently. By investing in the supply chain for these underlying technologies, Nvidia addresses a genuine strategic need whilst maintaining looser coupling with any single AI company. The logic is sound: the winners in the next phase of AI will be those with reliable access to the hardest-to-source components.
Critics might argue that Huang is being overly conservative, that the $30 billion investment is inadequate relative to OpenAI's ambitions, or that stepping back from the infrastructure deal signals doubts about OpenAI's long-term viability. The evidence does not really support this reading. Huang has made clear his continued belief in OpenAI as "one of the most consequential companies of our time." What has changed is not his conviction but his assessment of how to structure a relationship that serves both companies' interests whilst standing up to scrutiny.
There is also a harder fiscal case embedded here. Nvidia's shareholder base has begun asking uncomfortable questions about whether the company's massive investments in AI startups represent sound capital allocation or whether they are, in effect, a form of vendor financing that props up customer valuations. These are not frivolous concerns. During the dotcom boom, companies that engaged in similar circular arrangements often ended up writing off spectacular losses when the market corrected. Huang, by all accounts, has studied that history.
The question of whether $30 billion is the right number, or whether the photonics investments are properly sized, or whether Nvidia should have held the line on the original $100 billion plan is ultimately one reasonable people can dispute. What matters more is that Nvidia is applying some discipline to how it deploys capital in an ecosystem where discipline has been in short supply. That the company is simultaneously investing heavily in supply-chain diversification and resilience, including deliberately bolstering US-based manufacturing at Lumentum and Coherent, suggests a company thinking hard about geopolitical realities and long-term competitive positioning.
The messy truth is that Nvidia's pullback does not invalidate OpenAI's potential, nor does it vindicate those who have warned that AI valuations are entirely detached from reality. It simply reflects that even within the most exciting technological moment of the decade, prudent investors occasionally need to step back and ask whether the structure of a deal still makes sense. In this case, Huang decided it did not. That may disappoint those who celebrate grand ambition over careful planning, but it should reassure those who believe that the AI industry's credibility depends on sober management of resources alongside genuine innovation.