For more than two years, no trend has been held in higher regard on Wall Street than the evolution of artificial intelligence (AI). With AI, software and systems are capable of making split-second decisions, overseeing generative AI solutions, and training large language models (LLMs), all without the need for human oversight.
The long-term potential for this game-changing technology is truly jaw-dropping. If the analysts at PwC are correct, a combination of consumption-side effects and productivity improvements from AI will add $15.7 trillion to the global economy by the turn of the decade.
Although a long list of hardware and software/system application companies have benefited immensely from the AI revolution, none stands out more than tech titan Nvidia (NVDA 4.26%). But what you might be surprised to learn is that this highly influential AI company has scored a $4 billion "profit" in an uncharacteristic manner.
Image source: Nvidia.
Nvidia's hardware keeps AI data centers ticking
It took less than two years for Nvidia to catapult from a $360 billion market cap to (briefly) the world's largest public company, with a valuation that handily surpassed $3.5 trillion. A $3 trillion-plus increase in valuation in such a short time frame had never been witnessed before.
Nvidia's claim to fame is its Hopper (H100) and next-generation Blackwell graphics processing units (GPUs), which are the undisputed top options deployed in AI-accelerated data centers. Orders for both chips have been extensively backlogged, despite the efforts of world-leading chip fabrication company Taiwan Semiconductor Manufacturing to boost its chip-on-wafer-on-substrate monthly wafer capacity.
When demand for a good or service outstrips its supply, the law of supply and-demand states that prices will climb until demand tapers. Whereas direct rival Advanced Micro Devices was netting anywhere from $10,000 to $15,000 for its Instinct MI300X AI-accelerating chip early last year, Nvidia's Hopper chips were commanding a price point that topped $40,000. The ability to charge a premium for its AI hardware, due to a combination of strong demand and persistent AI-GPU scarcity, helped push Nvidia's gross margin into the 70% range.
Nvidia CEO Jensen Huang is also intent on keeping his company at the forefront of the innovative curve. He's aiming to bring a new advanced chip to market each year, with Blackwell Ultra (2025), Vera Rubin (2026), and Vera Rubin Ultra (2027) set to follow in the path of Hopper and Blackwell. In other words, it doesn't appear as if Nvidia will cede its compute advantages anytime soon.
The final piece of the puzzle for Nvidia has been its CUDA software platform. This is what assists developers in maximizing the compute abilities of their Nvidia GPUs, as well as aids with building/training LLMs. CUDA has played a pivotal role in keeping clients loyal to Nvidia's ecosystem of products and services.
Collectively, Nvidia's data center segment has helped catapult sales by 383% between fiscal 2023 (ended in late January 2023) and fiscal 2025, and sent adjusted net income skyrocketing from $8.4 billion to $74.3 billion over the same timeline.
Wall Street's AI darling just scored a $4 billion "windfall"
As you can imagine, most of Nvidia's more than $74 billion in adjusted net income last year was derived from its operating activities -- and this is how it should be for a market-leading growth stock. But it's not the only way Wall Street's AI darling can put dollars in the profit column.
What's often overlooked about Nvidia is that it's also an investor. Just as institutional money managers with more than $100 million in assets under management (AUM) are required to file Form 13F no later than 45 days following the end to a quarter -- a 13F lays out which stocks, exchange-traded funds (ETFs), and select options were purchased and sold -- businesses with north of $100 million in AUM must do the same. This includes Nvidia.
At the end of March, Nvidia had more than $1.1 billion invested across a half-dozen publicly traded companies. Accounting rules require Nvidia to recognize unrealized gains and losses each quarter, based on the change in value of the securities in its investment portfolio.
Nvidia's largest investment holding is AI-data center infrastructure goliath CoreWeave (CRWV -7.63%), which went public in late March. Nvidia made an initial investment in CoreWeave of $100 million in April 2023, and upped its stake by another $250 million in March 2025, prior to its initial public offering (IPO). On a combined basis, Nvidia has put $350 million of its capital to work in Wall Street's hottest IPO.
As of the closing bell on Friday, June 20, the 24,182,460 shares of CoreWeave that Nvidia held, as of March 31, were worth (drumroll) close to $4.44 billion. On an unrealized basis, Wall Street's AI titan is sitting on a $4 billion-plus "profit" from its investment.
If you're wondering why "profit" is in quotations, it's because Nvidia may have reduced its stake in CoreWeave since the second quarter began. We won't know for sure until 13Fs detailing second-quarter trading activity are filed in mid-August. Further, this $4 billion unrealized gain can fluctuate, depending on where CoreWeave stock closes out the June quarter. Nevertheless, it's been one heck of a windfall for Nvidia.
Image source: Getty Images.
Is Nvidia playing with fire?
While Nvidia has a solid track record of making smart investments in up-and-coming tech companies -- many of which it's partnered with -- there's also the real possibility it's playing with fire when it comes to CoreWeave.
Don't get me wrong, CoreWeave has been a fantastic client for Nvidia. It purchased 250,000 Hopper GPUs for its AI-data centers, which it leases out to businesses looking for compute capacity. It's in Nvidia's best interest that CoreWeave succeed and upgrade its AI chips roughly twice per decade.
But there are a number of red flags with CoreWeave that suggest its $88 billion valuation isn't sustainable.
One of the biggest concerns with Wall Street's hottest IPO is that Nvidia's aggressive innovation cycles could hinder, not help, its business. Bringing an advanced AI chip to market annually has the potential to quickly depreciate CoreWeave's Hopper GPUs, and might send customers to rival data centers that have newer chips. When CoreWeave looks to upgrade its infrastructure in the coming years, there's a very good chance it'll recoup far less from its assets than it expects.
CoreWeave has also leaned on leverage to build out its AI-data center. Relying on debt to acquire GPUs can lead to burdensome debt-servicing costs. For the moment, these servicing costs are adding to the company's steep operating losses.
Valuation is another clear concern with CoreWeave. Investors are paying roughly 8 times forecast sales in 2026 for a company that's not time-tested and hasn't generated a profit.
While Nvidia, undoubtedly, wants to see CoreWeave succeed, locking in its gains at these levels would make a lot of sense.
