Let's talk about whether NVIDIA is spending all the money it's earned recklessly.

Over the past few years, NVIDIA has demonstrated an extremely strong cash-generating capability amid the AI wave. According to disclosures as of 2025-10-26 (FY2026 Q3 10-Q), the company's book value of cash, cash equivalents, and marketable securities totals approximately $60.6 billion. Having a lot of cash is not scary; what's scary is "spending it wrong" after having a lot of cash.
Charlie Munger has a very poignant saying: once a company has too much cash, management easily falls into a state of "having money means you have to spend it," ultimately turning a business that could compound into a form of exhausting capital allocation. When cash piles up, the most common "disease" is wanting to turn the company into a bigger empire: larger budgets, larger departments, and a larger footprint. The problem is—scale does not equal returns. Many expansions look busy but are essentially diluting ROIC. Especially when the market is hot and valuations are expensive, the temptation to "buy stories, buy growth, buy synergy illusions" is strongest: even when the core business is strong, they end up doing a bunch of unrelated things, often with very good-sounding reasons—a second growth curve, synergies, strategy... The result is that the management span is stretched to the breaking point, efficiency declines, capital returns worsen, and shareholders end up footing the bill in the end.
So the question arises: Are NVIDIA's recent moves a plus or a minus for shareholders?
I'll state my core judgment upfront: NVIDIA's greatest strength has never been just "how many more cards it sold this year," but that it is building a platform—hardware + network interconnect + system + software stack + developer ecosystem. For this type of platform company, money should be spent on two types of things:
• Expanding the future profit pool: R&D investment, ecosystem building, filling key capability gaps
• Consolidating platform control: Supply chain/interconnect/hardware-software synergy, raising customer switching costs, making it harder for customers to leave
Jensen Huang has also repeatedly emphasized that much of the company's investment is centered around the ecosystem and the CUDA system. Translating that into shareholder language: Cash should be prioritized to feed the moat, not to start new ventures just to "look proactive." Based on currently disclosed information, I believe NVIDIA's overall direction is still moving towards platformization, at least not in a state of "having too much cash forces us to find something to do."

First, let's talk about a frequently overlooked but most practical "way to spend money": buybacks
Many people only focus on who it invested in or bought, but for shareholders, the most direct and easiest-to-measure form of capital allocation is buybacks and dividends.
In the first nine months of FY2026, NVIDIA disclosed share repurchases of approximately $36.7 billion, with cash dividends of about $0.732 billion during the same period, totaling roughly $37.4 billion. This at least shows that: it hasn't used all its cash to "tell stories"; a significant portion of cash is clearly returned to shareholders. Furthermore, the company's board added a large buyback authorization in FY2026 Q3, demonstrating strong capital discipline.

Next, let's discuss a few of the most watched moves
1) Groq: More of a smarter structure for "acquiring capability + reducing risk" than an "acquisition"
The market once simplistically summarized the Groq matter as "NVIDIA acquiring Groq." But a more accurate description is: NVIDIA and Groq reached a non-exclusive inference technology licensing arrangement, while Groq's founder and some core members joined NVIDIA, with Groq continuing to operate as an independent entity.
From a shareholder perspective, this structure is often more cost-effective than a direct merger:
• It can acquire key capabilities on the inference side (especially in the trend of increasing weight from "training→inference")
• Simultaneously reduces systemic risks common in M&A, such as regulatory, integration, and cultural conflict risks
• Non-exclusivity also means: NVIDIA is getting "capability and talent," not being forced to swallow all the baggage of a complete organization
In a nutshell: This is more like "getting the key puzzle piece in hand" rather than buying a company just for scale.
2) Intel: Rather than conspiracy theories, return to the logic of "platform binding"
Regarding NVIDIA's investment in Intel, there are various interpretations in the market, with some even simplifying it to "political factors." I'm more inclined to view it within the platform strategy: using equity to further bind the collaborative relationship, making CPU × GPU × interconnect into a more tightly coupled system-level solution.
The cooperation framework disclosed by Intel is very clear: both parties will work around NVLink interconnect for next-generation data center and PC products—on the data center side, Intel will build custom x86 CPUs for NVIDIA, which NVIDIA will integrate into its AI infrastructure platform and bring to market; on the PC side, it involves deeper integration of NVIDIA RTX GPU chiplets into x86 SoCs.
From this perspective, this investment seems more like an extension serving "platformization + ecosystem entry," rather than management chasing a "second curve" disconnected from the core business.
Additionally, the importance of this investment can also be seen from the disclosure approach: NVIDIA disclosed its Intel position in the 13F as of 2025-12-31 (approximately 215 million shares, with a market value of about $7.9 billion). This means it is no longer a symbolic small position but a clear bet written into the asset allocation map.
3) OpenAI: The piece where "boundaries" should be watched most closely
The arrangements related to OpenAI are the most controversial part and indeed the area where boundaries need to be watched most closely in NVIDIA's capital allocation. Because it could push NVIDIA from being a "shovel seller" towards being a "gold mine shareholder" for a stretch. If done right, the potential returns are higher; but if the single bet is too heavy, the risk is also amplified.
Here, my attitude towards it is more "conditional":
• If the deal ultimately goes through, it could form a kind of "capital recycling ecosystem": OpenAI, after raising funds, will likely continue to expand computing power purchases, and NVIDIA, as a core hardware and system supplier, will continue to benefit.
• But it also brings real risks: high valuation, governance and regulatory uncertainty, and a fast-changing competitive landscape. Once the cycle reverses or policies change, the drawdown could be ugly.
• More critically: This kind of super-large equity bet can easily escalate risk from "operational fluctuation" to a "structural variable." And structural variables are often not something that can be quickly corrected by quarterly earnings reports.
So, for myself, to judge whether NVIDIA is "spending money recklessly," I would watch three things:

Is the money being used to deepen the moat, or is it going towards empire expansion unrelated to the main course?
Is the "cash discipline towards shareholders" through buybacks/dividends still present?
Is there any sign of "overconcentration" starting to appear in any "massive equity bets"?
Overall, I believe NVIDIA's current spending direction remains clear: focusing on platformization and strengthening the CUDA/ecosystem moat. Moves like Groq and Intel can logically be linked back to "platform capability enhancement." What truly requires vigilance is bets like OpenAI—"super-large, super-high valuation, super-uncertain"—they might smell great, but they also most easily escalate risk from a controllable operational variable to an uncontrollable structural variable.