US President Trump said in a post on Truth Social: “I made $45 billion for the United States in 8 months!” The image card attached to the post also read: “Trump’s Intel investment is now up $45 billion!” The image shows that Trump bought Intel at a price of about $20, while Intel’s share price has risen to about $97. In the center of the card, it even highlights “45B” in huge lettering to emphasize that this investment has generated a $45 billion paper gain.
This post has drawn attention not only because Trump directly credited himself with the rise in Intel’s stock price, but also because Intel—once, for many years, a traditional semiconductor giant that the market briefly viewed as having “missed the AI wave”—is now returning to the market spotlight amid multiple factors, including a reassessment of the AI supply chain, a rebound in CPU demand, and tight supply of advanced packaging capacity.
Rising agentic AI, with CPUs returning to the core of data centers
In the past, the core narrative of generative AI was heavily concentrated on GPUs. Training and inferencing large language models rely on massive parallel computation, making NVIDIA the most prominent beneficiary in the AI era—and leading the market at one point to believe that the role of CPUs in AI infrastructure was becoming marginalized.
But when AI applications move beyond simply generating text and images and further toward “agentic AI,” computing demand starts to change. Agent systems don’t just answer a question once; they need to break down tasks, call tools, read data, repeatedly reason, and execute multi-step workflows. Workloads like these involve heavy data movement, multi-task coordination, system scheduling, and sequential computation—precisely the area CPUs have long been good at.
NVIDIA has also noted that the number of tokens generated during the operation of agentic AI systems grows exponentially, making “performance per watt” an important consideration for building data center hardware. As enterprises begin deploying AI agents that can run for long periods and continuously carry out tasks, data centers no longer just need more GPUs; they also need more high-performance CPUs to handle the coordination and execution workloads behind agent workflows.
This is why CPUs are being repriced by the market again. Bank of America estimates that the CPU market size could grow from $27 billion in 2025 to $60 billion by 2030. Both AMD and Intel are facing tight supply, with some products seeing lead times as long as six months and price increases of more than 10%. Analysts point out that wafer capacity constraints are the main cause of this supply crisis, and overall supply and demand may not improve meaningfully until 2026.
This is also the first layer of background for Intel’s stock rebound: AI is not just a GPU story. As AI infra shifts from model training to agentic deployment, CPU demand is being reopened.
Advanced packaging becomes the second main storyline: Intel’s EMIB is being seen again
The second storyline behind Intel’s comeback narrative is advanced packaging.
EMIB, short for Embedded Multi-die Interconnect Bridge, is Intel’s embedded multi-die interconnect bridge technology. Unlike traditional 2.5D packaging that relies on large silicon interposers, EMIB connects multiple dies or chiplets through small silicon bridges embedded in the packaging substrate. Intel argues that this approach reduces the use of additional silicon area, improves yields, lowers power consumption and cost, and also makes it easier to integrate chips from different process nodes and different IP blocks within a single package.
Analyst Jeff Pu says Intel’s EMIB yield has reached 90%, which is an important positive for Intel Foundry and helps explain why market confidence in Intel Foundry has recently improved. The report also mentions that Google’s next-generation TPU is rumored to adopt Intel’s advanced packaging, NVIDIA’s next-generation Feynman chips have also been linked in market chatter to EMIB technology, and Meta has been singled out as potentially using EMIB in a CPU program in the late part of 2028.
This suggests Intel’s opportunity may not be to immediately and positively challenge TSMC at the leading-edge manufacturing node, but rather to regain a position starting from the advanced packaging link that the AI supply chain lacks the most.
Citrini Research’s prior bullish core rationale for Intel is also advanced packaging. Citrini believes that the market has often simplified AI semiconductor competition into NVIDIA versus ASICs, TSMC versus Intel, or Blackwell versus TPU—but this framework ignores a deeper bottleneck: regardless of which AI chip ultimately wins, advanced packaging is still required first.
Google TPU, Amazon Trainium, Meta MTIA, and even future custom chips OpenAI might develop—all inherently move toward architectures with multiple dies, multiple chiplets, and multiple HBM. These chips are not fully interchangeable with one another; instead, they jointly consume limited advanced packaging capacity.
Therefore, Citrini argues that Intel’s opportunity is not to regain leadership in leading-edge processes in the short term, but to capture spillover AI packaging demand that emerges after TSMC’s CoWoS supply becomes oversubscribed, using EMIB and Foveros. In other words, front-end chip manufacturing may still be handled by TSMC or Samsung, but the final stage entering Intel’s advanced packaging process will allow Intel to regain a key position in the AI supply chain.
90% yield is a positive, but it’s still 8 percentage points away from a mass-production benchmark
However, Intel’s advanced packaging comeback narrative is not without risk. Analyst Guo Ming-chi points out that Intel already has experience with stable EMIB production, so the verified yield for the EMIB-T technology in development reaching 90% is a “positive but reasonable” signal. But internally, Intel uses FCBGA as the comparative yield benchmark for EMIB production, and the industry’s current FCBGA production yield is around 98% or higher.
This means that even though Intel EMIB-T has already crossed an important technical validation threshold, increasing from 90% further to 98% may be more difficult than going from concept to 90%.
On the surface, 90% and 98% differ by only 8 percentage points, but for high-priced, large-area, multi-die packaging products like AI chips, the yield gap will directly translate into costs, lead times, and effective output. Especially since Google’s next-generation TPU Humufish still has some specifications not yet finalized, and the verified yield in technology validation is not the same as the final yield in mass production. Therefore, while Guo Ming-chi looks positively at Intel’s long-term advanced packaging development, he also reminds that Intel’s ability to overcome mass-production challenges still needs to be watched in the near and mid term.
In other words, Intel’s comeback story is already being priced in by the market, but the real test is not whether it can make EMIB-T—it’s whether it can deliver stable mass production at the costs, yields, lead times, and scale demanded by AI customers.
This article Trump brags about his investment in Intel (Intel) multiplying fourfold: “I helped the U.S. make $45 billion in 8 months” first appeared on Lianxin ABMedia.
