Intel INTC 12.84%|Google 3M TPU Order vs. Nvidia Test Gap

The Same Week Intel Fell 11% and Jumped 13%

Intel's week between June 6 and June 9 compressed years of narrative into four trading sessions. On Friday June 6, Intel fell 11%. The May jobs report showed 172,000 positions added, nearly double the 80,000 consensus estimate. The 10-year Treasury yield surged past 4.54% and rate-hike odds climbed above 50%. Long-duration technology assets sold hard. Intel did not just decline — it led the chip sector lower alongside AMD. The Philadelphia Semiconductor Index dropped more than 5% in a single session. Intel was already carrying a complicated story before that Friday arrived. The foundry business generated $4.5 billion in revenue last quarter but remained in operating loss. The stock had climbed 182% year-to-date on turnaround expectations, not confirmed execution. When the macro environment turned hostile, Intel was exposed precisely because expectation had run ahead of delivery. Any stock priced on future execution is the first to reprice on rate uncertainty. Then Monday June 9 arrived with two separate reports from two different sources. Intel shares opened sharply higher and closed up 12.84% at $111.90. That move recaptured the $110 level and pushed through it. For a holder who watched the Friday slide, Monday's reversal was not reassuring. It was disorienting. The same stock, the same foundry in operating loss, the same unresolved questions about yield and profitability. Different price by 24 percentage points over 72 hours. What changed was not Intel's fundamentals. What changed was the external validation signal attached to its manufacturing node. Two separate customers moved on Intel 18A — one with a confirmed production order, one with a technical evaluation. Each carried a different implication for how seriously to take the foundry's progress. The gap between those two signals is exactly where the analysis lives.

Google Orders 3 Million TPUs From Intel — What That Number Means

The first signal came from Alphabet's Google. According to reports published June 9, Google committed to ordering more than 3 million Tensor Processing Units from Intel for delivery in 2028. These are not commodity chips. TPUs are Google's proprietary AI accelerators, built to power Gemini AI models and Google Cloud infrastructure. Google historically manufactured TPUs exclusively at TSMC. Moving a portion of that production to Intel signals that TSMC's supply constraints have become severe enough to force active diversification. TSMC's chief executive has publicly acknowledged that AI chip demand will exceed the company's supply capacity for years. Morgan Stanley analysts estimate Google's total TPU production across 2027 and 2028 could reach 6 million units or more. If Intel captures the reported 3 million, that represents roughly half of Google's projected two-year manufacturing need. Intel's entire foundry business generated $4.5 billion in revenue in its most recent quarter. A 3 million-unit TPU commitment running through 2028 is a meaningful volume anchor against that baseline. The node being used is Intel 18A. Intel has positioned 18A as the most advanced semiconductor process currently shipping in the United States. It incorporates gate-all-around transistors under the RibbonFET brand name and backside power delivery. Google's decision to qualify and commit at 18A scale means the node passed months of technical validation before the order was placed. Last week, Mizuho raised its Intel price target from $124 to $128 while maintaining a Neutral rating. The firm cited demand for agentic AI workloads that require CPU orchestration as the primary driver. Wells Fargo raised its target from $85 to $110. Neither firm upgraded Intel to Buy. That hesitation carries information. The analysts see the customer validation but are not yet convinced that yields and execution will sustain the foundry thesis through 2028. The buried assumption in the bullish read is that Google's 3 million unit order translates directly to foundry profitability on that node. It does not, automatically. Volume at 18A and profitable volume at 18A are two different conditions. Intel still needs yield rates and unit economics to improve before the foundry exits operating loss. The gap between customer commitment and confirmed profitability is precisely where Intel's execution risk concentrates.

Nvidia's 18A Evaluation — The Non-Order That Still Shifts the Landscape

The second signal from June 9 was structurally different from the Google order. Reports indicated Nvidia is evaluating whether Intel's 18A manufacturing process can produce a forthcoming chip that combines four graphics processing units into a single package. The testing uses multi-project wafers, which is the standard industry practice for qualifying a new manufacturing partner before committing to volume production. Nvidia has not placed a firm order with Intel. That distinction is critical and is where mainstream coverage has been imprecise. A multi-project wafer test is a technical qualification exercise, not a purchase agreement. Nvidia remains primarily reliant on TSMC for all AI accelerator production. Any shift to Intel would represent one of the most significant foundry realignments in semiconductor industry history. But the fact that Nvidia is running technical evaluations at all changes the competitive context regardless of outcome. Nvidia's data center GPUs, which generated over $75 billion in revenue last quarter, are TSMC-dependent. If TSMC's capacity constraints force Nvidia to seek alternatives at any volume level, Intel is the only domestically based foundry operating at the leading edge. Intel's Embedded Multi-die Interconnect Bridge and Foveros advanced packaging technologies are key differentiators in any evaluation Nvidia would run. For US-based hyperscalers with government-linked workloads, Intel's domestic manufacturing footprint also provides geopolitical supply chain diversification that TSMC's Taiwan operations cannot offer. Intel received a $5 billion co-investment from Nvidia and $7.86 billion in CHIPS Act funding — both anchoring the relationship before this evaluation began. The position Intel occupies after June 9 is unusual. It is not a confirmed Nvidia manufacturing partner. But it has crossed the threshold of being seriously considered by the world's most valuable chip company for production-scale work. For a holder sitting on Intel at $111.90, the question is whether testing converts to a firm order before the foundry posts its first profitable quarter. The timeline is long. Evaluations in 2026 map to volume decisions in 2027 at earliest. A holder who anchors conviction on Nvidia's test — rather than Google's confirmed order — is pricing an outcome that has not yet occurred.

Agentic AI and the CPU Bottleneck — Intel's Independent Revenue Vector

Beyond the foundry story, a separate structural argument for Intel is building in the analyst community. The shift from conversational AI to agentic AI is creating a CPU demand surge that was not priced into Intel's recovery thesis six months ago. Earlier AI workloads were GPU-centric. Training large language models required parallel computation at massive scale, and Intel's x86 CPUs were largely irrelevant to that race. Agentic AI operates differently. Agentic systems plan, reason, and execute multi-step tasks autonomously. They require a CPU to orchestrate control flow, manage memory allocation, and coordinate between processing elements. The GPU handles acceleration, but the CPU handles the logic layer. Industry forecasts cited in recent analyst notes suggest CPU requirements for AI workloads could increase three to four times by 2027. Intel's Core Ultra series, manufactured on Intel 18A in Chandler, Arizona, targets AI PCs and edge deployments directly in that demand window. Clearwater Forest, the forthcoming Xeon 6-plus server processor also built on 18A, targets enterprise servers running agentic pipelines in the cloud and on-premise. Zacks cited both products specifically in its June 8 analysis, noting that Intel's vertically integrated model combined with the Nvidia co-investment and CHIPS Act funding positions it as a direct structural beneficiary of the agentic shift. Mizuho's upgrade rationale named agentic AI explicitly as the driver behind the price target increase. Intel's client computing group generated $32.2 billion in revenue for all of 2025. If agentic workloads drive even a 20% uplift in CPU unit demand through 2027, the revenue impact is immediate and does not depend on foundry profitability. This is what separates the current Intel thesis from the 2023 and 2024 turnaround stories. A pure foundry thesis is binary — either 18A delivers at scale or the stock rerates lower. An agentic CPU thesis runs in parallel and has a shorter time-to-revenue path through the product division. The risk in Intel's position is executing both simultaneously under the same capital allocation constraints. The case for Intel this week is not simply that Google ordered chips. It is that three separate demand vectors converged on the same company in the same week. A hyperscaler committed to foundry capacity. The world's most valuable chip company began evaluating that same node. And a structural CPU demand shift independent of both is already reflected in sell-side price target upgrades. Whether the 182% year-to-date gain has already captured all three vectors, or whether the market is still calibrating, is the question a holder cannot answer without watching one specific signal. The forward checkpoint is Nvidia's foundry evaluation. If that converts from multi-project wafer testing to a firm production order in the next two quarters, the standing read on Intel shifts from cautious validation to confirmed structural inflection. If it does not, Google's 3 million TPU order remains meaningful but insufficient to justify a foundry-leadership multiple on a business still running at an operating loss.