The artificial intelligence boom has created shortages of chips, electricity, and data center space.
Billions of dollars are now being spent by Tesla (TSLA) and Nvidia (NVDA) on a project that most investors have probably never heard of.
Tesla and Nvidia will give Samsung Electro-Mechanics 3.7 trillion won, or $2.8 billion, to expand production of advanced semiconductor substrates for artificial intelligence chips, according to Reuters, citing Maeil Business Newspaper.
According to the article, Nvidia would pay around 2.6 trillion won and Tesla would contribute about 1.1 trillion won.
A flip-chip ball grid array, or FC-BGA, is the part at the core of the expenditure. It is not nearly as well-known as the graphics processing units made by Nvidia. However, such chips cannot function without it.
FC-BGA substrates facilitate the reliable operation of larger, increasingly complex chips by connecting sophisticated processors to the larger circuit board and delivering power and electrical signals.
These substrates become increasingly challenging to produce as AI computers grow in size and power. They therefore become yet another possible bottleneck in an AI supply chain that is already having difficulty keeping up.
Nvidia is paying to secure a hidden part of its chips
There is more to Nvidia’s engagement than just a declared investment.
On Oct. 21, Samsung Electro-Mechanics and Nvidia separately announced a strategic alliance to increase South Korea’s capability for producing specialized substrates.
The businesses want to build new manufacturing lines in Sejong and Busan, with Samsung providing substrates for Nvidia’s computer chips and networking hardware that connects data center CPUs.
By 2028, Nvidia may make up a sizable share of Samsung’s whole package-substrate manufacturing capacity. Samsung anticipates that the volume supplied to Nvidia will increase at a double-digit yearly rate.
For something that most customers will never see, this is an impressive commitment.
However, Nvidia clarified why it is important.
“The pace of the AI infrastructure buildout requires us to expand manufacturing capacity on a global scale,” Nvidia Executive VP of Operations Debora Shoquist said.
The supply chain for artificial intelligence is only as quick as its slowest link. Although Nvidia can create faster processors, they still need sophisticated substrates, networking hardware, memory, and packaging before they can function as systems.
According to Samsung, the latest substrates are becoming bigger and more intricate because AI processors need more power and electrical connections. Those production constraints limit the number of businesses that can consistently produce them.
Because of this shortage, Nvidia has an incentive to contribute to the funding of more capacity instead of just hoping vendors would someday provide more.
Nvidia’s $35 billion China signal comes with a catch.JOHN MACDOUGALL / Getty Images
Tesla is chasing the same AI supply chain
The narrative is made more intriguing by Tesla’s reported involvement.
The business is becoming more and more reliant on computation that goes far beyond the processors found in traditional automobiles.
Tesla is developing its Optimus humanoid robot, Robotaxi, Cybercab, and artificial intelligence infrastructure for autonomous driving. As a result, Elon Musk’s business is increasingly vying with the biggest IT firms in the world for some of the same resources used in chip manufacture.
More Tech:
Meta and Microsoft just sent employees a memo about Anthropic
Microsoft cutting off popular software with no extension Oct. 13
Apple’s biggest AI advantage may face new privacy challenge
Under the reported agreement, Tesla will spend more than 1.1 trillion won in the new Sejong manufacturing lines of Samsung Electro-Mechanics, according to Reuters.
The Tesla section of the statement announcing its relationship with Nvidia is now based on the Maeil Business article that Reuters quoted, since Samsung Electro-Mechanics has not officially disclosed the Tesla investment.
That difference is important. However, Samsung’s business already reflects the wider shift in demand.
According to the company, operational profit more than quadrupled to 440.4 billion won, while second-quarter revenues increased 24% from the same period last year to 3.46 trillion won.
Samsung attributed a portion of such gain to higher sales of high-value semiconductor substrates for server CPUs and artificial intelligence accelerators.
Samsung promised an even bigger expansion in September, pledging 2.51 trillion won to boost production in Vietnam and 4.27 trillion won to enhance FC-BGA capacity at its Sejong factory.
Samsung called the Sejong project its largest investment ever in a single product. That’s a lot of money pursuing something that most investors didn’t really care about previously.
AI spending is spreading beyond the famous chipmakers
That could be the most important investing lesson.
Nvidia’s graphics processors became well-known during the early stages of the artificial intelligence boom.
Exposing all of the less obvious infrastructure needed to deploy those processors by the millions is the next stage.
One limitation was high-bandwidth memory. Another was electricity. Additionally, advanced packaging has become a significant bottleneck. Substrates are currently drawing commitments totaling billions of dollars.
Only a few vendors are now able to produce the most advanced versions at the size and quality needed for next-generation CPUs, according to Samsung Electro-Mechanics.
The business said during its second-quarter results report that although clients are increasingly looking for long-term agreements and even offering financial assistance for new factories, bigger AI processors and higher usage of high-bandwidth memory are demanding more substrate capacity.
This situation contributes to the explanation of Nvidia’s apparent willingness to directly fund increased supply.
The business is no longer only purchasing parts. It’s becoming more necessary to ensure that the factories manufacturing these parts really exist.
As its aspirations expand into robots, driverless cars, and artificial intelligence, Tesla seems to be facing a similar issue.
Investors may see both possibilities and threats as a result of the AI surge. If one hidden part that holds a CPU together fails, a $100 billion data center cannot function.
The $2.8 billion amount is significant because of its potential impact.
Tesla and Nvidia are not investing that money on yet another ostentatious consumer product. They are helping secure anything hidden underneath the chips.
And it could be the most obvious indication yet that the next AI barrier might be concealed in areas that investors don’t often check.
Related: Nvidia has a $35 billion problem most companies would want