Altman's Global Chip Hunt
Although OpenAI is not the only tech giant lacking chips, Altman's visible urgency is certainly among the highest.
According to a senior manager at TSMC, Altman has been approaching various chip giants, including TSMC, since 2023 to discuss possibilities of increasing production capacity.
However, most executives did not agree with this plan, unwilling to invest large amounts of capital and specialized labor just to increase production for OpenAI alone. The risk was considered too high.
Last month, when TSMC CEO CC Wei was asked about the prospects of new factories, he said Altman's plan was "too aggressive and hard to believe."
But this was just the public statement. According to a senior TSMC manager, company executives privately promised Altman that if he personally or OpenAI could commit to large orders of new chips, they would be willing to expand chip production.
This arrangement was more feasible for TSMC as it could use existing factories rather than risking money on dozens of new plants.
Besides trying to stockpile more chips from suppliers, Altman is also simultaneously pushing for the establishment of new data centers.
According to an industry executive, Altman plans to cooperate with external investors to set up a new company specifically responsible for building data centers, which OpenAI would then rent servers from.
Insiders say Altman has sought approval from the U.S. Department of Commerce for this plan and has been negotiating cooperation with governments of other countries (such as in the Middle East).
Several governments, including the UAE, have declared their desire to develop AI-focused data centers to promote economic development, enhance their country's status in the tech industry, and protect local data from leaks and external influences.
Additionally, obtaining funds from private equity firms and other investors is also within Altman's consideration.
For instance, he has even discussed a separate "Altman project" with SoftBank, the Japanese company that invested $3 billion in Perplexity.
The Wall Street Journal reported months ago that Altman was negotiating with relevant investors, aiming to raise a staggering total of $7 trillion!
This figure even exceeds the total national debt of some major economies.
According to data from the Securities Industry and Financial Markets Association, the total bond issuance of all U.S. companies last year was $1.44 trillion. The combined market value of Microsoft and Apple, the two highest-valued U.S. companies, is only about $6 trillion.
Altman privately explained the composition of this astronomical figure: $7 trillion includes the total investment over several years for this project in real estate, electricity, and chip manufacturing.
Currently, it's unclear whether Altman has successfully secured investment or begun registering companies.
Altman himself rarely discusses the specifics of data center and chip plans publicly. But he privately told this industry executive that he would announce the plan this year.
According to insiders, this plan caused Altman considerable trouble last year, even being one of the triggers for the "dismissal crisis" at the end of the year.
At that time, his idea was to have the data center project exist independently from OpenAI, but the board was concerned that Altman's increasing investments and side projects would distract his attention from OpenAI.
However, after the "palace intrigue", this plan now faces almost no resistance within OpenAI.
Earlier this year, Altman informed colleagues that OpenAI would hold shares in the new project, and the new board has begun considering this proposal.
Challenging NVIDIA?
According to The Information's report, a person involved in Altman's conversations said that OpenAI's chip team will be led by Richard Ho, and they are likely to choose an American company to co-develop the chip, with Broadcom being the most likely candidate.
After all, Broadcom has previously worked with Google to manufacture TPUs and has already communicated with OpenAI's chip team.
The chip team's helmsman, Richard Ho, just joined OpenAI in November last year, presumably talent specifically recruited by Altman for the "chip-making" plan.
Richard Ho graduated from the University of Manchester in the UK with a bachelor's degree in Microelectronic Engineering, and later went to Stanford for a Ph.D. in Computer Science.
From 2014, he joined Google's hardware and semiconductor team as a senior engineer, participating in the TPU project, and gradually rose to become Chief Engineer and Senior Engineering Director until leaving in 2022.
Insiders say the OpenAI team doesn't seem to have started designing chips yet, with formal production not expected until at least 2026.
Furthermore, the development team is considering various memory component suppliers and potential chip packaging solutions to optimize performance.
For instance, HBM chips, one of the most important components of GPUs, directly affect the computational performance of chips.
HBM is responsible for providing high-speed connections between processors and memory, accelerating data transfer and processing, thereby speeding up AI computations.
These memory chips are in short supply just like GPUs, with demand consistently outstripping supply.
Insiders reveal that Altman has already discussed his chip plans with Samsung and SK Hynix earlier this year, both of which are top-tier HBM chip manufacturers.
Negotiating directly with upstream suppliers could significantly impact NVIDIA's business, and it's hard to say how NVIDIA might react.
But the fact is, not just OpenAI, but companies like Meta, Microsoft, Google, and Amazon are also heavily investing in developing their own AI accelerators.
These AI giants are both NVIDIA's biggest customers and are becoming its most powerful competitors.