Google’s 3 Million TPU Order Validates Intel’s Foundry Ambitions

Intel’s long-struggling foundry business received a substantial credibility boost on June 8, 2026. Multiple reports confirmed that Google placed a firm order for more than 3 million Tensor Processing Units (TPUs), set for delivery in 2028, following months of testing Intel’s advanced packaging technology. Concurrently, reports surfaced that Nvidia is actively testing Intel’s 18A process node and EMIB advanced packaging for a new GPU architecture expected the same year . Intel’s stock surged roughly 9–12% in early trading on the news .

These developments point to a crucial shift: large hyperscale AI customers are actively seeking alternatives to an overstretched TSMC, and Intel is increasingly seen as a viable second source for advanced AI silicon manufacturing .

The Google TPU Order: A Hard-won Credibility Milestone

The reported Google deal is not a tentative exploration—it is a firm production order for over 3 million TPUs in 2028. The order matters not only for its scale but for the technical validation it represents. Google spent months testing Intel’s advanced packaging before placing the order, meaning Intel passed meaningful technical scrutiny beyond symbolic engagement . Morgan Stanley estimates Google will produce more than 6 million TPUs across 2027 and 2028 combined, suggesting the Intel relationship could scale well beyond the initial 2028 allocation .

The primary driver, according to reports, is TSMC’s inability to meet surging demand for advanced manufacturing capacity. Google’s decision to formalize a second-source relationship with Intel signals that the industry’s dependency on a single dominant supplier is no longer tenable .

Nvidia’s Testing of Intel 18A and EMIB: Interest Without Commitment

Nvidia’s posture toward Intel is more cautious. Reports describe Nvidia as testing Intel’s 18A process node and EMIB packaging for its “Feynman” GPU architecture expected in 2028, but these reports characterize the activity as evaluation and early trials, not a confirmed production purchase order .

Earlier reporting from DigiTimes and other outlets suggests that even if Nvidia formalizes an engagement, it would likely assign only the I/O die—up to roughly 25% of the Feynman wafer content—to Intel’s 18A or future 14A node, while keeping the primary GPU compute die at TSMC . Such a partial win would still be meaningful. It would place Intel directly in the manufacturing flow for a leading AI GPU supplier for the first time, initially through lower-risk components rather than the flagship compute die .

Market Reaction: A Credibility Rally

Intel shares surged 9–12% on the morning of June 8, 2026, as the Google and Nvidia reports circulated. The rally reflected renewed investor optimism that Intel’s foundry strategy is attracting named, high-volume AI customers, moving beyond a pure turnaround narrative into tangible deal flow .

The Broader Intel-Google Partnership and Xeon CPUs

Beyond the TPU foundry order, Intel’s relationship with Google extends deeply into CPU infrastructure. In April 2026, the two companies announced a multiyear collaboration to continue deploying Intel Xeon processors in Google Cloud infrastructure, including for AI inference workloads. Google Cloud’s C4 and N4 instances already run on Intel’s latest Xeon 6 processors .

The partnership also involves co-developing custom Infrastructure Processing Units (IPUs) to accelerate networking, storage, and data center orchestration tasks. This CPU-focused relationship underscores that Intel’s strategy is not solely foundry-dependent; the company is positioning its Xeon lineup as a core component of AI infrastructure even as GPUs dominate the conversation .

CEO Lip-Bu Tan’s Turnaround Narrative: Foundry, CPUs, and the AI Era

Since taking over as Intel CEO in March 2025, Lip-Bu Tan has aggressively refocused the company. He cut management layers, sold non-core assets, and secured billion-dollar investments from Nvidia and SoftBank. A scheduled $8.9 billion U.S. government grant was converted into a federal equity stake .

On the foundry front, Tan has reported that Intel’s 18A process node is in volume production supporting the Panther Lake ramp, with yields improving at roughly 7% per month and tracking ahead of internal targets. The next-generation 14A node is moving forward with customer engagement, design kits, and a path toward risk production in 2028 and volume production in 2029 .

Tan has also made a deliberate argument that the next phase of AI will shift demand back toward CPUs. He notes that inference workloads surpassed training workloads in the second half of 2025, and that emerging agentic AI systems are considerably more CPU-intensive than traditional training environments. In his framing, the CPU becomes the orchestration layer and critical control plane for the entire AI stack, not merely a legacy component .

Intel’s turnaround remains a long-term effort with acknowledged supply constraints and manufacturing bottlenecks, but the convergence of a firm hyperscaler TPU order, expanding Xeon AI infrastructure deals, and the CEO’s disciplined narrative suggests the company is building a credible foundation as both a foundry supplier and a renewed force in AI compute.