By Silicon Star
NVIDIA is doing one thing: using its massive cash reserves to invest in companies that buy its own GPUs, then watching those companies use that money to buy even more GPUs.
It's like one hand pours water to create foam, while the other hand pours soap into it.
As of 2026, this chip giant has made investment commitments exceeding $40 billion in just five months, covering every layer from optical fiber manufacturing and data center operations to foundational model R&D. Its identity is shifting from a chip supplier to the most important capital allocator in the entire AI industry chain.
Some money is used to build things; some money is used to make things appear more valuable. NVIDIA is doing both.
A Multi-Billion Dollar Sprinkler, Drenching the AI Supply Chain
NVIDIA's investment moves this year have become so dense that they can be compared to a "sprinkler system."
Media describes its rhythm like an automatic irrigation system: turning direction every few weeks, spotting any parched patch of land and leaving a $2 billion investment: Synopsys in December, CoreWeave in January, Lumentum and Coherent on the same day, Nebius in March, followed closely by Marvell.
NVIDIA's money-sprinkling pace is almost mechanical, resembling less an investment portfolio strategy and more a supplier list from the procurement department.
The movement last week was even bigger.
NVIDIA reached an agreement with Corning, committing up to $3.2 billion to help the 175-year-old glass manufacturer build three new optical technology plants in the US, boosting production capacity for optical connectivity for AI infrastructure by 10 times and expanding optical fiber production by over 50%.
The next day, it granted data center operator IREN warrants worth up to $2.1 billion to cooperate on deploying 5 gigawatts of AI infrastructure.
Adding to that, its largest single bet of $30 billion on OpenAI, its investments in Anthropic and the now SpaceX-merged xAI, NVIDIA has completed at least seven listed company investments and participated in about 24 rounds of private financing this year, forming an investment matrix that fully covers "chips – optical communications – data centers – large models."
Jensen Huang's explanation sounds almost humble: "There are so many excellent foundational model companies. We are not picking winners; we want to support everyone."
But a glance at the financial reports reveals that this sprinkler system is irrigating precisely the demand chain for NVIDIA's own GPUs.
The invested companies are almost all major buyers of NVIDIA chips. The investment actions directly stimulate purchases of optical modules, GPUs, and data center infrastructure, which are then rented out as compute power to hyperscale customers like Microsoft, Meta, and OpenAI. The outline of a "self-operated demand loop" is already clearly emerging.
The Left-Hand-to-Right-Hand Cycle, and Intel's Paper Wealth
This is precisely where the controversy is heating up.
Goldman Sachs analyst Schneider used the term "circular revenue" in his report: NVIDIA's equity financing for OpenAI increasingly resembles a supplier funding its own customer, only for the customer to hand the payment back in the form of GPU purchases. It's like a food supplier investing in a failing restaurant, requiring the restaurant to keep buying its ingredients with the money.
In 2026, NVIDIA is expected to recognize approximately $13 billion in revenue from OpenAI alone. A large portion of these gross profits is then reinvested back into its investment in OpenAI. Some media commented on this, saying the AI boom has spawned a circular, creative financing feast where part of the money is just going in circles.
Critics compare these maneuvers to "vendor financing" during the dot-com bubble: equipment providers offer financial support to customers, who then use that money to buy equipment, making it increasingly difficult to distinguish whether the end demand is real. Once the market turns, the false demand can be exposed rapidly within a few quarters.
But Jensen Huang scoffs at the "circular financing" notion. He bluntly responded with "absurd" in an interview with Bloomberg, insisting that NVIDIA's investments constitute only a small fraction of the massive funds these companies need. He views them as votes of confidence in long-term, generational businesses, not financial engineering.
The most conspicuous paper return right now comes from Intel. At the end of 2025, NVIDIA acquired about 215 million new Intel shares for $5 billion; in the following eight months, Intel's stock price soared nearly sixfold, pushing NVIDIA's unrealized gains toward the $25 billion range.
Wedbush Securities analyst Bryson pointed out in a report that NVIDIA's dense transactions "fit neatly into the theme of circular investment." However, he noted that if executed well, these investments are also building a "supply chain moat" for NVIDIA that competitors cannot easily replicate.
Mizuho chip analyst Klein sees the investments flowing into optical communications, fiber optics, and silicon photonics components as a "highly sophisticated" use of funds by the CFO team, capable of accelerating the development of bottleneck technologies within NVIDIA's supply chain. But regarding the funds flowing into new cloud providers like CoreWeave and Nebius, he was more candid: "These investments make me a bit uncomfortable, like paying for your own GPUs in advance."
When the cycle turns, the market will eventually ask: how much of the current AI demand on the books is genuine, and how much is merely a mirage propped up by NVIDIA's own balance sheet?
The Money Isn't Hot, But Confidence Can Shatter First
There are many technical metrics to measure this high-stakes gamble, but nothing is more direct than NVIDIA's own financial reports.
In Q4 FY2026, NVIDIA's revenue was $68.1 billion, a 73% year-over-year increase. Annual revenue reached $215.9 billion, up 65%. Annual net profit was $120.067 billion. Free cash flow was a staggering $97 billion. The company returned $41.1 billion to shareholders through stock buybacks and cash dividends.
Yet, the day after releasing its strongest-ever earnings report, NVIDIA's stock fell about 5.5%, wiping out roughly $260 billion in market value in a single day. It wasn't that the report was bad; it was that "exceeding expectations" itself had lost its marginal impact.
The market is fixated on the 2-nanometer chip fab Terafab, on optical interconnects, on the new Rubin architecture, only to find that the only certainty available for purchase is the next 12 months. How much longer can AI capital expenditures expand? How far can the growth duration be stretched?
NVIDIA certainly isn't feeling the heat; its cash reserves are enough to keep it as the central water tower of the AI era for a long time. The money isn't in a hurry, but confidence can shatter first. Confidence shatters at the corner of circular investment. When a supplier must use its own capital to irrigate its customers' demand, the market starts calculating how much natural precipitation is still beneath this lawn.
A bubble isn't necessarily a fraud. Sometimes it's just all the participants in the world believing in the same thing simultaneously. But when you keep adding soap to the bubble, making it look sturdier, more lasting, and more like the steel-and-concrete engineering itself, it becomes an engineering experiment requiring extremely cold judgment from investors.
