Comment Intel And Tesla Partner With Terafab. But How?

Intel wants to support Elon Musk in his ambitious Terafab project. Two tech giants side by side? Shareholders will be delighted. However, nothing is known about the strategy behind the strategic partnership. But perhaps that IS the strategy.

Intel is now also on board Elon Musk's much-heralded Terafab project. The company announced on Tuesday that it is now part of the Tesla/SpaceX initiative. "Our ability to design, manufacture and package extremely powerful chips at scale will help accelerate Terafab's goal of generating 1 TW of computing power per year to drive future advances in AI and robotics," Intel CEO Lip-Bu Tan was quoted as saying in a statement from the company. This could help Tesla, SpaceX and xAI to secure sufficient computing power for next-generation AI and robotics applications. Terafab represents "a fundamental shift in how silicon logic, memory and packaging will be manufactured in the future."

Fine Words, Little Substance

These are all fine-sounding words. However, nothing is known about what form this support will take or over what timeframe. Is Intel providing foundry capacity? Is Intel actually directly involved in the design of the chips in demand? How is the goal of producing both logic and memory chips in a single plant to be implemented technologically? Didn't Tesla only recently sign a long-term manufacturing deal worth 16.5 billion US dollars with Samsung to have these chips manufactured in the new plants in Texas? And above all, when is this legendary Terafab supposed to be built by?

The following information about the project is known so far: AI chips with an "output of 1 terawatt per year" are to be manufactured there. There was talk of 2 nm structures. The chips are to be "an order of magnitude" more powerful than "what is currently on the market". Analysts estimate the costs for this project at 5 trillion US dollars (yes, in English there is talk of five trillion dollars!). There is no information on when this is to be implemented or in which year the construction of such a plant is even planned. The construction of a "regular" fab can usually cost 30 billion US dollars and take a good three years before the first chips roll off the production line. And this applies to companies that have an idea of how to design and implement manufacturing processes and IP nodes.

Of course, there is now much speculation as to how Intel could be involved in this project. One sensible approach could be to work with Tesla to establish a coordinated supply chain with a capacity pool that includes both its own facilities and those of other chip manufacturers such as Intel, according to speculation e.g. on Tom's Hardware. However, this contradicts the claim that Terafab will develop and manufacture "everything in one place". There is also speculation about customized production facilities jointly financed by Intel and Tesla, or at least product lines. Or even that Intel could be commissioned entirely with design and production chips via the Intel Foundry Service. But then why the "own" fab, which is supposed to be bigger than anything that has been built in the chip manufacturing sector so far?

"Elon has proven many times that he can reinvent entire industries," wrote Intel CEO Lip-Bu Tan in a post on Musk's personal social media platform X. "This is exactly what is needed in semiconductor manufacturing today. Terafab represents a fundamental shift in how silicon logic, memory and packaging will be manufactured in the future." That sounds all well and good, but it doesn't answer the question of how or of any apparent strategy. And the fact that these words come from the CEO of a company that has recklessly squandered its former advantages in processor production over the last decade also has a certain aftertaste.

Lots of Broken Things, But Hardly Any Results

Honestly, what's left of Musk's big promises of recent years? What can the Grok AI model from xAI ultimately do better than the models from Anthropic, OpenAI or Google, apart from perhaps trivializing the Holocaust or producing deepfake porn of real people? How revolutionary were Tesla's Cybertrucks or the grandly announced e-trucks again? And how far along is Tesla with its self-driving vehicles? Elon has certainly moved fast and broken things, but what results are there now? "Move fast and break things" may be a great alternative slogan to "courage to fail", but it shouldn't refer to the speed at which a self-driving autopilot drives through a wall or human-sized dummies. When you consider the sums that have to be spent on building fabs and how crucial purity and cleanliness are in wafer production, this slogan makes me dizzy. Please try to imagine the credo "move fast and break things" in the face of a high-NA-EUV system from ASML.

Yes, it is quite possible to acquire your own chip know-how and develop great processors yourself at some point. Apple has demonstrated this well with the M series. But Apple has consistently built up this knowledge over almost two decades and didn't just pluck it out of the ground in a few years. And the M processors are still manufactured at TSMC.

Two Companies Chasing the Technology Leaders

Speaking of TSMC, this is where the real problem lies: As much as Intel is desperately looking for customers for its in-house 18A foundry process, the manufacturing specialists from Taiwan are still considered the measure of all things. Samsung's foundries are ahead when it comes to memory, and Sk Hynix or Micron are also outperforming Intel in this field in terms of technology.

From this point of view, you have to ask yourself why Tesla/SpaceX would want to partner with Intel in this area. One inevitably wonders whether a certain "Make America great..." I mean, "Made in America" tendency might not play a role here.

Of course, Musk has probably already gained an advantage: In principle, he probably now has the opportunity to look into the cards at Intel's processor production and Samsung's memory production. This should also be essential, as neither Tesla, SpaceX nor xAI have any expertise in process nodes or production technologies to date.

In the End, Only the Share Price Counts

But perhaps this is just two technological players entering into a pure publicity partnership in the hope of attracting customers and creating new confidence among investors. After all, it seems that the latter in particular is the real intention behind Elon Musk's plans, as the businessman is aiming for an unprecedented initial valuation of 1.75 trillion US dollars with the upcoming IPO of SpaceX! And Intel would also benefit from a confidence boost from its shareholders: the company recently bought back its shares in its Irish fab in Leixlip, but still has hardly any well-known customers for its manufacturing processes. Naturally, both companies now need promising-sounding and attention-grabbing partnerships.

Business Punk summarized this strategy wonderfully. "This is a clash of two desperate people: Musk needs manufacturing expertise that his companies don't have. Intel needs customers that no one else wants."

Who knows, maybe I'm just cynical. Maybe I'll be sitting at my desk in 2030, literally eating these words I've written. But it's more likely that a SpaceX rocket will land on Mars.(sg)

Tips & tricks for plastics professionals

Future trends in additive manufacturing