(March 24): When Elon Musk took the stage on Saturday to unveil his plans to get into semiconductor production, he didn’t spare the superlatives.
“I have an important announcement to make, which is the most epic chip-building exercise in history by far,” he said before a small crowd in Austin, Texas. “This is really going to take things to the next level, a level probably people aren’t even contemplating right now. We’re going to adjust the context by a few orders of magnitude here.”
No one has ever proposed anything quite like what Musk calls Terafab. The project, as outlined, would be a massive operation to build cutting-edge semiconductors for artificial intelligence, robotics and space forays. He would not only try to take on the best chip manufacturer in the world — Taiwan Semiconductor Manufacturing Co — but he also wants to do it at volumes far beyond the industry’s current capacity.
The scale boggles the mind. The project would require something like US$5 trillion to US$13 trillion in capital spending, according to estimates from Bernstein analysts. That would fund 140 to 360 new factories, each making 50,000 wafers per month in order to reach the one terawatt of annual computing capacity he proposed.
Musk, by far the richest person in the world, has accomplished what others believed impossible before — creating a commercially viable rocket business with SpaceX, bringing electric vehicles to the mainstream with Tesla Inc and delivering internet connectivity from space with Starlink. But some doubt Musk can, or even intends to, build what he sketched out in Austin.
See also: Taiwan's ASE to invest $12 mil in AEM via placement at $3.591 each
“A true Terafab feels like a stretch to us,” wrote the Bernstein analysts, including Stacy Rasgon. The amount of compute would be “on the order of the entire current global installed semi capacity and would in fact require many multiples of current installed capacity for ‘relevant’ semis.” Patrick Moorhead of Moor Insights & Strategy wrote that it’s unlikely Musk will ultimately build chip fab facilities at all.
Rather, Musk’s goal with his Terafab proclamation may be something different: to highlight the growing shortage of chip production capacity or motivate chipmakers to step up. Or it may boost the prospects for SpaceX as it heads for an initial public offering later this year.
There are growing concerns in Silicon Valley that the semiconductor industry isn’t ramping up fast enough to produce the chips AI companies will need to meet their ambitious business plans. Amazon.com Inc, Alphabet Inc and other hyperscalers expect to spend about US$650 billion this year alone to build out data centre infrastructure. That’s already creating a severe crunch in memory chips and is beginning to spill over into AI accelerators.
See also: ASM’s sales outlook beats estimates as China demand rebounds
Nvidia Corp’s Jensen Huang has talked about how he’d like more out of TSMC, the Taiwanese firm that makes all of his advanced semiconductors. But the linchpin of global chipmaking has taken a disciplined approach to expansion. “We directly receive very strong signals from our customers’ customers, requesting the capacity to support their business,” CEO C C Wei told analysts during an October earnings call. “We will also remain disciplined and thorough in our capacity planning approach to ensure we deliver profitable growth for our shareholders.”
Musk was explicit about how Tesla, xAI and SpaceX are going to need enormous quantities of silicon in the years ahead. He wants to eventually have one terawatt of computing power for AI services and hundreds of millions of new Optimus humanoid robots every year. The “key missing ingredient” is computing power, he said, estimating the current AI output is only about 2% of what his companies need.
He said he has urged suppliers including TSMC, Samsung Electronics Co and Micron Technology Inc to expand as quickly as they can — and promised to take pretty much everything they make.
“We will buy all of their chips; I have said these exact words to them,” he said.
But semiconductor companies have opted to expand at a more measured rate, one far short of what he thinks necessary for Tesla, SpaceX and xAI.
“So we either build the Terafab or we don’t have the chips,” he said. “And we need the chips.”
Analysts are sceptical not just because Musk has never made chips before. He’s proposing to do it in a way that contravenes the economics of the chip sector. Building factories to supply your own needs was the way the industry got its start. But the steep price — a cutting-edge facility costs about US$30 billion and can be obsolete in as little as five years — means you have to produce and sell a massive number of chips to justify the fixed costs.
To stay ahead of Singapore and the region’s corporate and economic trends, click here for Latest Section
There’s logic to Musk’s recognition of the risks of supply shortages and what that might mean for his ambitions for the vehicle and robot markets, according to Barclays Capital.
“However, as is frequently the case with Tesla’s efforts, this endeavour is unprecedented in a number of ways,” its analysts wrote in a research note. “Tesla lacks experience in manufacturing chips, with significant cost and execution risk. Accordingly, a partnership with either Samsung or TSMC, or perhaps even Intel, could be a more likely route.”
Nvidia is just one example of a company that’s never owned a factory. The majority of its peers are in the same boat. They primarily outsource their production to TSMC and, to a lesser extent, Samsung.
Those companies are able to make the economics work by essentially pooling the business of most of the industry to get a volume of orders that pays for the capital expenditure and research and design.
Musk proposed that Terafab would use what’s known as the integrated device manufacturer model, which means designing and making chips in-house. That’s the model Intel Corp used to dominate for decades. But the company is now a shadow of its former self — in part because of missteps that highlight the enormous complexities of fabricating chips.
In-house chipmaking has faded since Morris Chang founded TSMC in 1987 and created the pure foundry business that freed up tech companies to focus on design. Today, the Taiwanese firm serves as the go-to chipmaker for Nvidia, Apple Inc, Advanced Micro Devices Inc, Broadcom Inc and Google — just to name a few.
In addition to a different business model, Musk also seemed to envision a very different manufacturing setup than the ones used by TSMC and Samsung.
“They’re getting clean rooms wrong in these modern fabs. I’m going to make a bet here that Tesla will have a two-nanometre fab and I can eat a cheeseburger and smoke a cigar in the fab,” he said in a podcast in January.
Clean rooms are critical for modern chip production. Individual transistors are many times smaller than a virus. Just one speck of dust can cause havoc and millions of dollars of wasted effort.
During Saturday’s livestream, Musk said the Austin fab will house equipment for logic semiconductors, memory chips, packaging, testing and lithography mask production in a single building. No existing chip firm has an operation like that. The specialisation and differences in processes mean it’s typically not economically feasible.
At the same time, Musk also appeared to be invoking US interests as a reason for his new chip venture. He retweeted a post on X saying that Terafab is vital to national security as the bulk of the world’s chips are still made in Taiwan.
To be sure, Musk’s semiconductor ambitions should be viewed through the prism of audacious ideas over the years: helping humankind colonise Mars, shooting people through sealed tunnels from New York to DC and reusing space rockets. At least one of those dreams is now reality.
Musk’s vision touched a nerve because it highlights one of the most pressing concerns facing the industry today: a shortage of the high-end semiconductors that companies and governments desperately need to deploy game-changing AI.
Musk “has admittedly done more than one thing that naysayers have called impossible at the outset", the Bernstein analysts wrote. “Perhaps Elon has something more off-the-wall in mind to improve things (we leave that as an exercise for the reader, as we don’t know what it would be).”
Uploaded by Arion Yeow
