Intel Reportedly Secures 3M TPU Order from Google
According to The Information, Alphabet’s Google has placed a large-scale order of more than 3 million Tensor Processing Units (TPUs) with Intel Corporation, with production expected to begin in 2028. The news reportedly triggered a sharp market reaction, pushing Intel’s shares up more than 13% in pre-market trading on Monday.
Citing sources familiar with negotiations between Google and Intel, the report indicates that the order involves Google’s in-house AI accelerator chips (TPUs) and reflects an exceptionally large deployment scale. The move comes amid rapidly rising global demand for AI compute capacity, which has tightened supply at leading foundry services providers such as TSMC, prompting major cloud and AI players to diversify their supply chains.
eMarketer analyst Jacob Bourne noted that the development highlights a broader industry trend, stating that major AI leaders are actively seeking supply chain diversification, while manufacturing remains heavily concentrated among a small number of advanced semiconductor producers.
For Intel, this agreement represents another significant win in its foundry and advanced manufacturing strategy. Since the leadership transition under CEO Lip-Bu Tan, Intel has reportedly secured multi-billion-dollar investments and strategic engagements from various technology and government-linked entities. Previous reports also indicated that Tesla, Inc. selected Intel as a key manufacturing partner for its next-generation 14A process, tied to its planned AI semiconductor production ecosystem in Austin, Texas.
Additionally, earlier reporting from The Wall Street Journal suggested preliminary agreements between Intel and Apple Inc. for partial chip manufacturing collaboration. Analyst Gil Luria of DA Davidson commented that beyond standard diversification needs, companies such as Google and NVIDIA have stronger incentives than ever to engage with Intel, as supporting Intel’s manufacturing capabilities aligns with broader U.S. industrial policy priorities.
The report also notes that NVIDIA has been evaluating Intel’s ability to produce advanced multi-die GPU architectures integrating four graphics compute tiles into a single package, although no formal orders have been confirmed.
Intel declined to comment on the report, while Google and NVIDIA have not issued immediate responses.
However, the report does not clearly specify whether the 3 million TPU order is related to wafer fabrication or advanced packaging services. Industry analysis suggests that the engagement is more likely tied to advanced packaging rather than full wafer production. Earlier in May, Morgan Stanley reported that due to tight CoWoS advanced packaging capacity at TSMC, Google’s upcoming “Humufish” TPU program—designed by MediaTek—may leverage Intel’s EMIB packaging technology, with mass production targeted for 2027.
Although concerns remain regarding the scalability of EMIB for high-volume ASIC production, recent supply chain checks suggest Intel’s EMIB yield has reached above 90%, with supporting ecosystems such as substrates and silicon capacitors now in place.
“Humufish” refers to Google’s ninth-generation TPU project designed by MediaTek. At Google Cloud Next ’26, Google officially unveiled its eighth-generation TPUs, separating AI training and inference workloads into two dedicated chips: TPU 8t for training and TPU 8i for inference, with the latter designed by MediaTek.
MediaTek CEO Rick Tsai previously stated that its first AI ASIC project for a major U.S. cloud customer is progressing well, expected to generate around US$2 billion in revenue by late 2026, with potential expansion into multi-billion-dollar scale by 2027. The company is also working on a second AI chip program targeting mass production by the end of 2027, alongside multiple data-center initiatives currently in advanced negotiation stages.
