OpenAI buying NVIDIA chips with NVIDIA investment… controversy over 'AI circular trading'

OpenAI signs AMD chip supply agreement

Secures AI chips at bargain prices by leveraging share price increase

NVIDIA invests US$100 billion in OpenAI

OpenAI leases NVIDIA GPUs in return

Altman: "The economic value of AI is limitless"

#NVIDIA is investing US$100 billion in OpenAI, and OpenAI is paying US$22.4 billion to CoreWeave, an AI data center lessor, to lease AI data centers. CoreWeave's spare computing power is then bought by NVIDIA for up to US$6.3 billion.

Such "circular transactions," in which AI infrastructure companies invest in AI developers and that money is then used to buy AI infrastructure, are drawing increasing controversy. Bloomberg reported on the 8th (local time) that "some analysts are skeptical about the circular transactions formed by AI giants."

The circular trading controversy was raised after NVIDIA announced on the 22nd of last month that it would invest US$100 billion in OpenAI, and it intensified when OpenAI and AMD reached a deal on the 6th.

Under this deal, OpenAI will receive a total of 6GW of AI accelerators from AMD. Of that, 1GW will be delivered next year in the second half as AMD's MI450 product. In exchange, OpenAI can buy AMD common stock at US$0.01 per share. If AMD's share price rises to about US$600, roughly four times the trading price at the time, OpenAI would be able to purchase up to 160 million shares. By announcing a partnership with OpenAI and driving up AMD's share price, OpenAI would be able to secure AI chips at bargain prices through that price increase.

The previously announced NVIDIA investment in OpenAI also has a strong circular transaction characteristic. If NVIDIA invests US$100 billion in OpenAI, most of that investment money will be used again to lease NVIDIA GPUs. In addition, NVIDIA has invested in Intel (US$5 billion) and xAI (US$2 billion), and OpenAI has signed a computing power supply contract with Oracle worth US$300 billion.

These circular transactions can occur only if there is an expectation that the AI models, which sit at the apex of an AI ecosystem composed of GPUs-AI servers-cloud-AI models, will generate enormous future profits. The problem is that the timing of monetization for AI developers like OpenAI cannot be predicted. OpenAI is the world's largest startup, valued at US$500 billion, but it is still running at a loss. According to The Information, OpenAI's revenue for the first half of this year was US$4.3 billion (about KRW 6.1 trillion), and its loss was US$7.8 billion (about KRW 11 trillion). Jensen Huang, NVIDIA's CEO, also said on the 8th, "(OpenAI) still has no money," and added, "Whether through equity or debt, it will have to raise funds through exponentially increasing revenue."

However, OpenAI and NVIDIA plan to further expand such transactions. OpenAI CEO Sam Altman said that day, "You can expect us to show much more in the coming months," and emphasized, "If AI models develop in the directions we anticipate, the economic value is limitless." CEO Huang also argued, "This is fundamentally different from the dot-com bubble of the 2000s," and said, "A new industrial revolution is beginning."

Silicon Valley = Kim In-yeop, correspondent inside@hankyung.com