Nvidia’s $100B AI Gambit: Brilliant Empire-Building or Dot-Com Déjà Vu?

Massive OpenAI Partnership: In late September 2025 Nvidia (NVDA) announced it will invest up to $100 billion in OpenAI, supplying AI chips to support OpenAI’s next-generation models ^[1]. The deal calls for OpenAI to deploy at least 10 gigawatts of Nvidia systems (millions of GPUs) for its data centers, with Nvidia progressively funding $100B as each gigawatt comes online ^[2].
GPU Sales Lock-In: Nvidia’s investment is structured so that the cash essentially goes back into buying its own hardware. OpenAI’s CEO Sam Altman noted “everything starts with compute,” and said Nvidia was the only partner “that can do this at this scale”. In effect, Nvidia is financing a customer to buy more Nvidia GPUs – a model analysts call “circular financing” ^[3] ^[4].
Data-Center Deals: Nvidia already has a 7% stake in CoreWeave, an AI-cloud startup, which now represents ~91% of Nvidia’s known equity portfolio ^[5]. In September 2025 NVDA signed a new $6.3 billion order with CoreWeave, agreeing to purchase any unsold cloud-computing capacity through 2032 ^[6] ^[7]. This backstop protects CoreWeave’s customers and cements Nvidia’s role in their infrastructure.
Other Investments: Beyond OpenAI and CoreWeave, Nvidia has poured cash into many AI infrastructure projects. For example, it invested $700 million in UK-based data-center builder Nscale (part of a $1.1 billion funding round with Dell, Nokia, etc.) ^[8], and spent over $900 million to acquire AI networking startup Enfabrica’s CEO, team and technology ^[9]. The company is also involved in other “AI factory” data-center efforts and government-backed AI infrastructure (e.g. in the UK) as part of a broader ecosystem strategy.
Market Impact: These moves have helped fuel explosive demand for Nvidia’s GPUs and sent related stocks soaring. NVDA remains the world’s most valuable company with a $4+ trillion market cap ^[10] ^[11]. Its GPUs are in such high demand that analysts at Deutsche Bank and others say AI infrastructure spending could hit $4 trillion over the next few years ^[12] ^[13].
Critics’ Concerns: Some experts caution Nvidia’s strategy resembles Cisco’s late-1990s playbook. Cisco famously financed telecoms to buy its routers – boosting short-term sales but eventually resulting in $900 million of bad debts after the bubble burst ^[14] ^[15]. Likewise, analysts warn that Nvidia is essentially funding its own future GPU sales, which “on paper helps secure demand” but could backfire if demand cools ^[16] ^[17]. Concerns about antitrust and market distortions have been raised as well ^[18] ^[19].

The $100B Bet on OpenAI

On September 22, 2025 Nvidia and OpenAI unveiled a historic infrastructure partnership. Under the agreement, OpenAI will deploy 10 gigawatts of Nvidia’s latest AI systems – roughly 4–5 million GPUs – in new data centers for training and inference ^[20]. Nvidia will progressively fund up to $100 billion of this build-out, providing cash injections as each gigawatt of capacity is completed ^[21]. In practice, this means Nvidia puts up the capital for OpenAI’s growth, which is then spent on Nvidia’s own chips and platforms.

OpenAI’s CEO Sam Altman hailed the deal: “There’s no partner but NVIDIA that can do this at this kind of scale, at this kind of speed,” he said in a joint announcement. Jensen Huang, Nvidia’s CEO, called it “the biggest AI infrastructure project in history”. Altman emphasized that massive compute resources are essential to future AI breakthroughs: “This is the fuel that we need to drive improvement, drive better models,” he explained. (On Nvidia’s side, Jensen Huang noted the collaboration builds on a decade of partnership: he hand-delivered Nvidia’s first DGX system to OpenAI in 2016 and joked this deal is “a billion times more computational power” than that original box.)

Aside from headlines, details from Nvidia’s filings reveal that the initial tranche will be $10 billion (when the first gigawatt is built) and further investments come “in stages” as new data centers go live ^[22]. Oracle’s cloud is also involved via a $300 billion cloud deal, meaning much of OpenAI’s hardware runs on Oracle infrastructure ^[23]. Analysts note the strategic value: by financing OpenAI’s expansion, Nvidia effectively locks OpenAI into its GPU ecosystem. As Bernstein’s Stacy Rasgon put it, Nvidia is “now helping finance one of its biggest customers to keep buying its chips” – a textbook case of “circular financing” ^[24]. Rasgon cautioned that this cycle “will fuel” concerns that Nvidia is merely creating future demand for itself ^[25].

Importantly, regulators will watch such deals. Reuters reports that rivals and antitrust watchdogs could see Nvidia’s investments as undermining competition ^[26]. On the other hand, Nvidia argues these investments benefit the entire industry by ensuring AI projects get the infrastructure they need. In any case, the deal cements Nvidia’s role at the center of the generative-AI gold rush, with OpenAI (a leading AI innovator) effectively on retainer via Nvidia’s funding and hardware.

Building AI Factories: CoreWeave, Nscale and the Ecosystem

Nvidia isn’t just backing OpenAI. It has been aggressively funding AI data-center startups to create a broad “AI infrastructure” ecosystem. The crown jewel here is CoreWeave, an AI-cloud company that rents Nvidia GPUs to customers. Nvidia took about a 7% stake in CoreWeave during its March 2025 IPO, and CoreWeave now accounts for roughly 91% of Nvidia’s known stock investments ^[27]. CoreWeave’s business depends almost entirely on Nvidia hardware – it runs specialized GPU data centers and relies on Nvidia for the chips that power them.

In a new move, Nvidia signed a $6.3 billion contract with CoreWeave. Under this pact, Nvidia will buy any unsold compute capacity in CoreWeave’s data centers through April 2032 ^[28] ^[29]. In effect, if CoreWeave has idle GPU resources (say customer demand dips), Nvidia steps in as the buyer. As Investopedia explained, “As part of a $6.3 billion deal, Nvidia is obligated to buy any unsold cloud-computing capacity through April 13, 2032” ^[30]. CoreWeave itself announced the deal in an SEC filing on Sept. 9, 2025 ^[31]. This arrangement removes key risk for CoreWeave (no stranded capacity) and signals Nvidia’s confidence in continued demand. Analysts note that Deutsche Bank even added CoreWeave to its “Buy” list, citing the Nvidia deal as a major positive.

The CoreWeave deal, combined with Nvidia’s equity stake, makes CoreWeave almost an “AI factory” built around Nvidia chips. As Motley Fool writes, this is a “fantastic” move for both companies: CoreWeave mitigates risk, and Nvidia bets on robust AI demand ^[32]. CoreWeave’s other customers include Microsoft, OpenAI and Meta, so Nvidia is indirectly tying itself to those companies’ AI pipelines as well ^[33].

Another example is Nscale, a UK-based hyperscaler aiming to build “AI factory” data centers across Europe. In September 2025 Nscale raised $1.1 billion in Series B funding led by Aker ASA and joined by Dell, Nokia, Nvidia, and others ^[34]. Nvidia’s involvement here – along with the UK government’s support – shows its interest in global AI infrastructure. CEO Josh Payne says the funding will accelerate building secure, energy-efficient data centers to meet “surging” AI demand ^[35]. Nscale has also partnered with OpenAI on UK AI infrastructure plans.

Nvidia’s investments extend further: it backed a new venture fund run by CoreWeave to support other AI startups ^[36], and it is generally encouraging a network of GPU-based clouds and services. Industry observers note that Nvidia is acting almost like a “central bank” or stimulus program for AI: when financiers want to build out huge data centers, Nvidia is often there to lend money, buy hardware capacity, or even provide loans. (One commentator on social media even quipped that “Nvidia even assumes the role of a central bank” for AI, backstopping large projects when financing is tight ^[37].) In short, Nvidia is using its enormous cash flow to seed and strengthen the entire AI ecosystem – which in turn drives more GPU sales.

Talent and Technology: The Enfabrica Acquisition

Beyond financing, Nvidia has spent lavishly to secure talent and technology. A standout example is Enfabrica, a startup developing advanced AI networking chips (sometimes called AI fabric or interconnect chips). In September 2025 media reports revealed Nvidia “dropped a cool $900 million” to hire Enfabrica’s CEO and key engineers, and to license its technology ^[38]. Effectively, Nvidia acquired Enfabrica via a cash-and-stock deal rather than a typical acquisition (similar to Meta and Google’s recent talent buys). The deal closed and Enfabrica’s CEO Rochan Sankar joined Nvidia, according to CNBC sources ^[39] ^[40].

Enfabrica’s tech – a high-speed fabric chip and software for connecting GPUs/CPUs – promises to solve scaling bottlenecks in AI data centers ^[41]. By bringing this expertise in-house, Nvidia aims to improve GPU utilization and reduce costs in massive AI clusters. The $900M price tag (over $260M was already invested in Enfabrica, including by Nvidia) highlights how fiercely Nvidia is competing for talent and IP in the AI hardware race.

This talent acquisition underscores a pattern: Nvidia is not just selling hardware; it’s buying into the infrastructure layer. It hired former Intel, Meta and Google staff for its own AI chip projects, and it plans custom AI chips like “Blackwell” and “GH200” GPUs. Paying nearly $1 billion to secure Enfabrica’s team shows Nvidia is willing to invest heavily in any missing pieces of the AI stack.

Critics Sound the Alarm: Cisco Parallels and “Circular” Growth

While Nvidia’s moves are celebrated by investors, some analysts and commentators are cautious. The biggest concern is the resemblance to Cisco Systems’ late-1990s strategy. During the dot-com era, Cisco aggressively financed telecom carriers through its Cisco Capital arm to buy Cisco routers and switches. This boosted Cisco’s revenue (financing deals accounted for ~10% of its $20B annual sales) ^[42]. But when the telecom bubble burst, many of those customers defaulted and Cisco had to reserve about $900 million for bad loans ^[43]. High-profile failures like Rhythms NetConnections saw Cisco ship $20M of gear on credit just before an IPO, only to be left with $30M unpaid after Rhythms went bankrupt ^[44].

Nvidia’s current strategy has led critics to coin the term “circular financing.” As the Motley Fool’s Geoffrey Seiler writes, Nvidia is essentially funding one of its largest customers (OpenAI) to keep buying Nvidia products ^[45]. In effect, Nvidia’s balance sheet is underwriting its own future demand. If AI demand ever slows, some worry Nvidia could be left holding unneeded inventory or bad debt. Seiler explicitly likens this to Cisco’s mistake: back in 1999 Cisco’s loans “were a sales machine” – great until “the capital dried up and the entire market collapsed” ^[46].

Bernstein’s Stacy Rasgon voiced this worry to Reuters: Nvidia’s $100B bet will surely stoke those “circular financing” concerns raised during the dot-com bubble ^[47]. After all, large parts of the deal money will flow right back to Nvidia hardware revenues. Even Goldman Sachs in 2023 warned that overly bullish forecasts for AI hardware could be “too optimistic” if they don’t account for such feedback loops (though Nvidia’s executives argue otherwise).

Another risk is regulatory and competitive. By tying OpenAI, CoreWeave, and other AI leaders so closely to its ecosystem, Nvidia may draw antitrust scrutiny. Rival chipmakers (like AMD or new startups) could claim the playing field is being tilted by Nvidia’s cash. Already, NVIDIA’s broadened role (chipmaker, infrastructure financier, and now partial “AI bank”) is unprecedented. Some question whether Nvidia can sustain this pace of spending if markets turn or if OpenAI decides to diversify (OpenAI is reportedly working on its own chips as well).

Lastly, market sentiment is at stake. Nvidia’s stock has tripled in 2025 even as valuations in other tech sectors lag. If Nvidia’s intricate deals fail to deliver expected growth, some analysts warn the stock could correct sharply. After all, Cisco’s stock never recovered its 2000 peak as those financing missteps became clear ^[48]. In sum, while Nvidia’s leadership radiates confidence, many experts advise caution and close scrutiny of how these massive initiatives evolve.

Why Nvidia Is Betting Big (and What It Faces)

Nvidia’s torrent of investments reflects both opportunity and threat. On one hand, its GPUs are the undisputed workhorse of today’s AI boom: from ChatGPT to Google’s Gemini, most top AI models are trained on Nvidia hardware. Every new breakthrough or user makes those GPUs more valuable. By financing partnerships across the ecosystem, Nvidia is driving that growth rather than passively selling chips. It gains insight into demand (partners give forecasts), locks in sales through deals, and raises barriers for rivals. As one Nvidia insider noted on LinkedIn, the strategy also serves to “make it harder for competing chips to enter the market” ^[49].

Furthermore, Nvidia argues that scaling AI is a global endeavor that requires coordination. By investing in data centers, talent and complementary tech, Nvidia ensures the whole “AI supply chain” has the legs to run. In this view, its $100B is a strategic commitment to build an industry, not an isolated gamble.

On the other hand, the pace of spending is dizzying. Nvidia’s Q2 FY2026 revenue was $46.7 billion (up 56% Y/Y) ^[50], meaning the $100B commitment is more than two years’ sales at today’s rate. That cash comes from hyper-profits driven by current AI demand. But will demand remain “insatiable” as one analyst put it ^[51]? If new competitors emerge or AI growth slows (for example, if an AI “winter” arrives or if customers develop their own chips like Google’s TPUs), Nvidia could find itself overextended.

In comparison to similar cases, Nvidia’s move is far larger than past tech partnerships. Even Cisco’s late-90s loans peaked at a few billion dollars; Nvidia’s multi-billion-per-deal spree is unprecedented in tech. If it works, Nvidia will have cemented a virtual monopoly over AI infrastructure. If it fails, the losses could be enormous (though Nvidia’s digital assets and collateral likely mitigate some risk).

The Road Ahead: Growth or Overheat?

For now, Nvidia is widely hailed for making the “smart” move by supporting the very customers and ecosystem that feed its growth ^[52] ^[53]. Its stock has soared, and many investors treat these deals as evidence of the company’s unassailable position in AI. Sam Altman’s quip about Nvidia doing what only it can underscores how uniquely situated Nvidia is.

Yet history urges prudence. Cisco’s story reminds us that financing your own customers can backfire if the market turns. Meta and Google didn’t cross-invest in each other so broadly; Nvidia’s web of mutual interests is unusual and risky. The coming years will test whether the AI boom justifies these bets. If demand keeps growing as Nvidia expects (the company predicts multi-trillion-dollar infrastructure spending ^[54]), Nvidia will likely profit immensely. If not, the criticism of a “house of cards” may prove prescient ^[55].

Sources: Nvidia and OpenAI announcements ^[56]; Reuters and CNBC reporting ^[57] ^[58]; Motley Fool/Nasdaq analyses ^[59] ^[60]; Investopedia on CoreWeave ^[61] ^[62]; Capacity Media on Nscale ^[63]; Los Angeles Times on Cisco ^[64] ^[65], among others. All insights and figures are up-to-date as of September 2025.