In short: Intel has signed on as the primary foundry partner for Elon Musk’s Terafab, a $25 billion joint venture between Tesla, SpaceX, and xAI targeting a terawatt of AI compute per year, handing the struggling chip giant the marquee customer it has been searching for since pivoting to a foundry-first strategy.
On 7 April 2026, Intel announced it is joining the Terafab project, becoming the foundry partner for the most ambitious semiconductor facility ever proposed in the United States. The announcement came two weeks after Musk first unveiled Terafab at the North Campus of Giga Texas in Austin, a joint venture between Tesla, SpaceX, and xAI that claims it will produce one terawatt of AI compute every year. Intel’s role is to contribute its most advanced process node, packaging expertise, and manufacturing scale to make that claim real. For Intel chief executive Lip-Bu Tan, who has spent the past year attempting to rebuild Intel around an external foundry business, the deal is the most significant external customer win the company has landed since he took the job.
What Terafab is claiming to build
Terafab is designed as a vertically integrated semiconductor complex, covering chip design, lithography, fabrication, memory production, advanced packaging, and testing under a single roof, with a stated goal of producing between 100 billion and 200 billion custom AI and memory chips per year. The initial buildout targets 100,000 wafer starts per month, with ambitions to eventually scale to one million wafer starts per month at full capacity. The project involves two separate facilities on the Giga Texas campus: one dedicated to chips for automotive and humanoid robotics applications, including Tesla’s Full Self-Driving system, its Cybercab robotaxi programme, and the Optimus robot line; and a second for high-performance AI data centre infrastructure and specialised processors for orbital deployments.
That orbital component is central to the project’s rationale. SpaceX, which completed its acquisition of xAI in an all-stock deal in February 2026, creating a combined entity valued at approximately $1.25 trillion, is building out a constellation of space-based AI satellites internally designated AI Sat Mini. Musk has said 80% of Terafab’s compute output will be directed toward that orbital infrastructure, with the remaining 20% for ground-based applications. The full cost of the project has been cited as between $20 billion and $25 billion, though independent analysts have been sharply sceptical of whether that figure is remotely sufficient to meet the stated production targets. A note from Bernstein Research estimated the true capital required to hit one terawatt of annual compute at approximately $5 trillion, more than 70% of the total annual United States federal budget.
Intel’s role, and what the deal is worth
Intel will contribute its 18A process node, the company’s most advanced logic manufacturing technology, currently ramping to high-volume production at Intel’s fabrication plants in Arizona and Oregon. Intel’s 18A is a 1.8-nanometre-class node, placing it in the same tier as the most advanced processes currently entering commercial production globally, and it represents the most sophisticated semiconductor capability manufactured entirely within the United States. Intel’s statement on joining Terafab was direct: “Intel is proud to join the Terafab project with SpaceX, xAI, and Tesla to help refactor silicon fab technology.” The company added: “Our ability to design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab’s aim to produce 1 TW/year of compute to power future advances in AI and robotics.”
Tan’s post on X was more personal in its framing. “Elon has a proven track record of reimagining entire industries,” he wrote. “This is exactly what is needed in semiconductor manufacturing today. Terafab represents a step change in how silicon logic, memory and packaging will get built in the future. Intel is proud to be a partner.” Intel’s shares rose approximately 4% on the announcement, closing at $52.91. The market reaction reflects how significant the deal is for Intel’s foundry ambitions: in its most recent full year, Intel Foundry generated just $307 million in external customer revenue, a figure that makes the company a distant also-ran against Taiwan Semiconductor Manufacturing Company, which generates tens of billions annually from external customers. Terafab, if even partially realised, would transform Intel Foundry’s commercial profile entirely.
Intel’s recovery, and what this bet requires
Tan inherited an Intel in acute crisis. The company had lost ground to TSMC and AMD across almost every major product category, its own manufacturing roadmap had slipped repeatedly, and its foundry business, the effort to manufacture chips for external customers as TSMC does, had attracted little meaningful interest beyond government-supported contracts under the US CHIPS and Science Act. Tan’s restructuring has been aggressive: thousands of redundancies, a sharper focus on Intel’s 18A and 14A process nodes as the foundation of the foundry pitch, and a deliberate effort to position Intel’s domestic manufacturing capability as a geopolitical differentiator at a moment when US policymakers are intensely focused on reducing dependence on Taiwanese chipmaking.
Terafab is the clearest expression yet of where that pitch lands. The CHIPS Act tailwinds, the Trump administration’s desire to see advanced semiconductor production in the United States, and the specific demand Musk’s companies represent for high-volume, US-manufactured chips at the leading edge, all of those forces converge in this partnership. Whether Intel’s 18A can deliver at the yields and volumes Terafab’s targets require is a separate question. The node has been in development for several years and is only now entering volume ramp; the gap between a controlled high-volume manufacturing ramp and the production scales Terafab envisions remains very large. Chipmakers building the largest foundries in the world require several years of construction and billions of dollars before the first wafer is processed. The scale of capital commitments now characterising AI infrastructure investment gives some context for what serious execution at Terafab’s claimed targets would actually require.
The credibility problem Terafab has not solved
The scepticism around Terafab is structural, not merely financial. Building a 2nm-class fabrication facility capable of 100,000 wafer starts per month costs roughly $25-35 billion on its own, according to Tom’s Hardware’s analysis of Bernstein’s research, meaning the entire stated Terafab budget is roughly enough to build a single fab operating at a fraction of the claimed full-capacity scale. Reaching one million wafer starts per month would require dozens of such facilities. The $20-25 billion figure appears to represent initial construction capital for the first phase, rather than the cost of the stated ambition.
There is also the question of the companies at the table. SpaceX-xAI’s internal situation has been turbulent: all 11 of xAI’s original co-founders have now left the company since the SpaceX acquisition, a rate of attrition that has raised questions about the organisation’s technical continuity. Musk’s companies have a documented history of announcing timelines for facilities and products that subsequently stretch by years. Tesla’s Cybertruck, Optimus, and Full Self-Driving have each missed multiple committed dates without affecting the company’s willingness to make new commitments. None of this disqualifies Terafab, Musk’s companies have also delivered on goals that were widely dismissed, most notably SpaceX’s orbital launch programme, but it establishes why analysts are not taking the one-terawatt headline at face value.
What the partnership means for the chip industry
Intel’s arrival at Terafab lands at a moment when the chip industry is navigating a broader restructuring of who makes what and for whom. The rise of custom AI silicon, Amazon’s Trainium, Google’s TPUs, Microsoft’s Maia, has been eating into the share of AI workloads that run on Nvidia hardware. Nvidia’s response has been to open its NVLink Fusion interconnect to third-party silicon, including Marvell’s custom AI accelerators, a strategy designed to keep custom chip buyers inside Nvidia’s ecosystem even as they move off pure Nvidia hardware. Terafab represents something different: a vertically integrated attempt to produce custom silicon at a scale that has no precedent outside of the established foundry giants. If the project proceeds anywhere near its stated ambitions, it would add a third major domestic US semiconductor manufacturing ecosystem to a landscape currently dominated by TSMC’s Arizona expansion and Samsung’s Texas operations.
For Intel, the strategic logic is clear. As hyperscalers and technology companies increasingly pilot non-Nvidia chips for AI training and inference workloads, the market for foundry services from a domestically situated, leading-edge manufacturer is growing precisely when Intel has positioned itself to serve it. Whether Terafab is the vehicle that finally validates that positioning, or another ambitious announcement that tests the distance between Musk’s projections and physical reality, will become clearer as construction begins and wafer starts are counted rather than promised. The capital flowing into AI infrastructure at this scale has a way of turning implausible timelines into achieved ones, and Intel, for the first time in years, is positioned to benefit if it does.
Kyle McGinley helped build an AI-integrated 
You must be logged in to post a comment Login