Google just placed one of the largest orders yet for its own AI accelerators with a supplier other than TSMC. The move, confirmed in recent days, pairs with an earlier infrastructure pact and hands Intel a badly needed vote of confidence.
According to The Motley Fool, Alphabet’s Google ordered more than three million of its tensor processing units from Intel for production in 2028. The chips will come from Intel’s foundry business after months of testing its advanced packaging technology. Short sentences. No hedging.
But the story runs deeper. Back in April, the two companies expanded a long-running partnership. Google committed to deploy multiple generations of Intel’s Xeon processors across its global cloud footprint. It also agreed to co-develop custom infrastructure processing units, or IPUs, that offload networking, storage and security tasks from the main CPUs. The April 9 announcement came directly from Intel’s newsroom.
CPUs regain ground in the AI stack
The timing matters. Hyperscalers once chased pure accelerator performance. Now they face bottlenecks elsewhere. Agentic workloads and large-scale inference demand more from general-purpose processors than many expected. Nvidia’s own infrastructure chief told CNBC in March that CPUs were becoming the constraint. Intel’s deal with Google underscores that point.
Lip-Bu Tan, Intel’s CEO, put it plainly in the joint release. “Scaling AI requires more than accelerators – it requires balanced systems. CPUs and IPUs are central to delivering the performance, efficiency and flexibility modern AI workloads demand.” Amin Vahdat, Google’s senior vice president and chief technologist for AI infrastructure, echoed the sentiment. “Intel has been a trusted partner for nearly two decades, and their Xeon roadmap gives us confidence that we can continue to meet the growing performance and efficiency demands of our workloads.” Both quotes appear in the Intel Newsroom release and were picked up by Reuters.
Google’s own numbers tell the efficiency tale. Its latest specialized TPUs, the 8t for training and 8i for inference, deliver two to four times the speed of prior generations while cutting costs by 30 percent. Vahdat explained the logic in comments relayed by The Motley Fool. In a world of AI agents, specialization for training and serving unlocks gains that generic hardware cannot match.
And yet Google still needs the surrounding silicon. Xeon processors handle training coordination, latency-sensitive inference and the everyday cloud workloads that keep data centers humming. The IPUs free up those CPUs to do more of what they do best. The combination, the companies say, improves utilization and total cost of ownership across Google’s fleet. No financial terms were disclosed. No volume commitments for the Xeon side appeared either.
The foundry angle changes everything for Intel. For years the company struggled to attract major external customers to its manufacturing arm. This order, reportedly half of Google’s planned TPU output for 2028, marks its largest AI-related foundry win to date. Shares reacted immediately. Intel climbed sharply in the days after The Information broke the news on June 8, with some sessions showing gains near 12 percent. Recent X chatter shows traders piling in, citing the Bank of America double upgrade from underperform straight to buy and a new price target of $135.
But questions linger. Intel still carries losses. Its trailing earnings remain negative even as the stock trades well above some older consensus targets. Execution on the 18A process node, the one slated for these TPUs, will decide whether the momentum holds. Google, for its part, gains a second source. Dependence on any single manufacturer at these volumes carries risk. Diversification suddenly looks prudent when demand forecasts keep rising faster than anyone projected.
The broader context favors both sides. Google’s cloud backlog nearly doubled quarter over quarter to more than $460 billion. That figure, highlighted in The Motley Fool analysis, reflects explosive enterprise adoption of its AI services. Selling access to its TPUs to select customers expands the addressable market further. Alphabet shares, recently trading around 25 to 28 times forward earnings, appear reasonable against 19 percent expected revenue growth.
Intel, meanwhile, stacks this Google win atop other recent announcements. Partnerships with Nvidia for potential backup manufacturing, involvement in Elon Musk’s Terafab project and government support have all fed the narrative of a turnaround. The Google TPU order provides the clearest third-party validation yet for its foundry ambitions.
Analysts caution that one deal does not remake a balance sheet. Still, the signal is unmistakable. In an industry racing toward ever-larger clusters, the companies that control their full stack from silicon to software hold structural advantages. Google owns the models, the cloud and now a scaled internal accelerator supply with Intel as manufacturing partner. Others rent the chip layer from Nvidia. That difference compounds over time.
So the April infrastructure pact laid groundwork. The June foundry order scales it. Together they lock in a multiyear relationship that touches both the heart of Google’s AI operations and Intel’s hopes for manufacturing resurgence. Short-term stock pops may fade. The underlying shift in how hyperscalers think about supply chains looks more permanent. And that is the part worth watching.
WebProNews is an iEntry Publication