Four customers, whose identities were kept anonymous for competitive reasons, directly purchased goods and services collectively worth 46% of Nvidia’s $30 billion in turnover. That amounts to roughly $13.8 billion, according to the company’s 10-Q regulatory filing published alongside its hotly anticipated quarterly investor update.

Each was responsible for more than a tenth of overall top line, and their purchases were all related to its booming business selling chips to data centers, the likes of which entrepreneurs such as Elon Musk are in a rush to build amid the gold rush in artificial intelligence. 

To put that into perspective, this quartet of customers contributed more in sales than Nvidia reported for the prior year’s period as a whole. 

Video of the inside of Cortex today, the giant new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI pic.twitter.com/DwJVUWUrb5

— Elon Musk (@elonmusk) August 26, 2024

Although the names of the mystery AI whales are not known, they are likely to include Amazon, Meta, Microsoft, Alphabet, OpenAI, or Tesla.

Nvidia’s hottest products are AI chips like the H200. These are needed to train large language models like OpenAI’s GPT-4. They are also used to power inference, the process that ChatGPT or Sora uses to generate answers to text-based prompts.

This dependency on a handful of major customers also highlights a rising concern in the market over just how sustainable this abrupt, exponential growth from just one corner of its business can be. Some investors like Elliott Management and Citadel have voiced skepticism over how long this can be maintained.

History does give reason for concern. The semiconductor industry is known for its boom-and-bust cycles.

Shares in Nvidia are expected to open lower on Thursday, underperforming the broader equity market.

One customer provided revenue greater than Nvidia’s second-largest business

In fact Nvidia’s business ties with these whales are so significant the company flags them in a section of its quarterly reports titled “concentration of revenue,” which is dedicated to cluster risks.

“We have experienced periods where we receive a significant amount of our revenue from a limited number of customers,” it stated in its 10-Q regulatory filing, “and this trend may continue.” 

It’s a trend that is staggeringly profitable. Nvidia pocketed $5.60 out of every $10 of revenue it made over the entire first half as net income—margins most companies can only dream of. 

That explains why profit after tax nearly quadrupled to $31.5 billion during this six-month period over the previous year. Whether revenue, and therefore earnings, can continue to grow at this blistering pace is crucial to its investment story.

Take “Customer B” cited in the filing, for example: Its direct purchases represented 11% of Nvidia’s $30 billion in revenue. That means a single company contributed more in business than the group’s second largest division—gaming, with $2.9 billion—did as a whole.

Customer B, however, remained below the 10% threshold for the entire first half, which suggests it significantly ramped up spending during the past quarter seemingly out of the blue. The exact same could be said about “Customer C”; the numbers provided by Nvidia are identical. 

$NVDA pic.twitter.com/Lym4uHBCXs

— Jesse Cohen (@JesseCohenInv) August 27, 2024

Speaking to Bloomberg TV on Wednesday, CEO Jensen Huang answered a question on where demand is coming from, beyond the handful of hyperscalers like Microsoft, Google, and Amazon.

“We’re relatively diversified today,” he claimed, citing a range of different customer groups. 

Yet his own company’s numbers appear to dispute that conclusion. This time last year, for example, there were no direct customers whose business made up 10% or more of total revenue—neither for the first nor the second quarter.

Nvidia didn’t immediately respond to a request from Fortune for comment.