Foundries & Fabrication

Anduril, a defense contractor, is moving into computing hardware with Qualcomm, targeting military buyers first, then first responders and consumers. The prototype is only about 32% US-origin by cost, so the 65% federal procurement threshold at launch is unconfirmed. Qualcomm plans to make the processors at TSMC Arizona, which adds a US-based demand source for the foundry's domestic capacity. The launch is funded by the two companies rather than by subsidies, and Luckey calls it likely a low-volume product for critical sectors. Execution risk is high because the display, memory and other component supply chains are not yet domestic.

The designation opens more federal financing for helium producers and brings closer government monitoring of a gas chipmakers cannot readily substitute. The US produced 43% of global helium in 2025, but extraction plants are at capacity and new ones take years to permit and build. Qatar's disrupted exports and Russia's export controls raise supply-security concerns for fabs and other helium-dependent users. Domestic helium producers and gas processors gain a policy tailwind, while semiconductor, MRI and quantum users gain supply-risk mitigation but not near-term relief. Helium is the 20th item on the list and only the second gas after fluorine, which signals that industrial gases are now treated as supply-chain chokepoints.

Microsoft's Maia 300 pushes the largest hyperscalers toward in-house silicon for inference, the fastest-growing AI compute workload. Microsoft claims 40 percent more tokens per megawatt than leading GPU-based systems, which would weaken the case for buying merchant GPUs for inference at Microsoft. Nvidia and AMD face a risk to a major customer's inference demand, though training demand and the 2027 timeline limit near-term effects. TSMC benefits from leading-edge N2P demand and from the advanced packaging that chiplet designs require. Power efficiency per token becomes the main competitive metric, which favors custom ASICs and the vendors that supply them.

The deal signals a structural push to localize advanced semiconductor packaging in the US, a step long seen as the weakest link in domestic AI chip manufacturing. GlobalFoundries gains a new high-value revenue stream and strategic relevance in leading-edge packaging alongside TSMC and Intel Foundry. Nvidia, AMD and other HBM-dependent chipmakers stand to benefit from reduced reliance on Taiwan round-trips once capacity comes online in 2028, though near-term bottlenecks persist. The announcement intensifies competitive dynamics among Amkor, Intel Foundry and GlobalFoundries for US CoWoS-compatible packaging business. Investors should view this as a multi-year infrastructure build rather than an immediate capacity relief, given the 2028 timeline.

This funding round signals accelerating investor and corporate appetite for alternatives to Nvidia's dominant position in AI inference silicon. Samsung's involvement is strategically significant, giving it a stake in a potential non-Nvidia chip architecture while leveraging its memory and foundry capabilities, which could open new manufacturing and supply relationships. The move adds to a growing list of hyperscaler and startup efforts (OpenAI's Jalapeño, Google, AWS, Meta) fragmenting the AI chip market away from single-vendor reliance. While Euclyd remains unproven at commercial scale until 2028, continued capital flows into inference alternatives could gradually erode Nvidia's pricing power and market share in inference workloads specifically, even as it retains dominance in training. Long-term, this reinforces a broader industry trend toward diversified AI silicon supply chains and custom architectures optimized for cost and energy efficiency.