The Intel Foundry Opportunity: Two Paths to Anchor Customer Scale
The bull case, or even the economic case to justify Intel Foundry does not require taking customers outright from TSMC. In a day and age where there is more demand than capacity, there is essentially enough demand to go around and we believe Intel is better positioned than Samsung as the second source of choice. The opportunity for Intel Foundry spans two complementary segments: custom AI accelerators and server CPUs. Both are being capacity-constrained at TSMC as the world’s leading foundry concentrates on AI infrastructure.
Why Intel?
The shortfall of capacity to meet demand at TSMC remains one of the most discusssed in the semiconductor industry. This challenge to meet demand in both manufacturing wafer capacity and advanced packaging (CoWoS) capacity is the biggest driver of TSMC’s continued growth upside. We recognize that TSMC is set to grow 30% or more in 2026, and this projection is only capped by their ability to bring new capacity online. The reality is TSMC could generate significantly more revenue if they only had capacity to meet that demand. That demand gap is the opportunity for other foundries (some will say it’s an opportunity for TSMC OSATs and that is true also).
TSMC’s capacity shortfall opens up the opportunity for other foundries like Intel and Samsung. It is our view that Intel is much better positioned to capitalize on this opportunity than Samsung for a few reasons. First, Intel is increasingly becoming a nationalized player with the current US administration rewarding companies committed to America First. This will extend benefits to companies that choose to be customers of Intel Foundry. It must not be understated the strategic national importance that the US have semiconductors made in the USA and made by a USA company.
The second dynamic in Intel’s favor is superior process technology and advanced packaging capabilities relative to Samsung. Intel’s foundry technologies are in some cases extremely competitive with TSMC. This double benefit of political favor (lower tariffs, better tax incentives, and other elements of favor from the US government) and a competitive process makes Intel much better strategically positioned than Samsung among the leading semiconductor customers in the market like Apple, Qualcomm, Broadcom, Marvell, and others. I am intentionally leaving out NVIDIA and AMD from this list because we do not believe they will leave TSMC for core GPU designs (maybe for CPU…). These companies are well supported by TSMC and do not face issues getting priority access to new nodes and advanced packaging as customers lower down the tier list do. On Samsung, we should also note they have not put out a competive process technology since their 5nm node if we are being generous. We think the opportunity for Intel is at least two paths, and we lay out what we feel is the most compelling in the short term.
The AI-ASIC Opportunity
With packaging becoming much more of a bottleneck than many realize, we believe Intel’s extremely competitive advanced packaging menu will remain attractive. TSMC’s CoWoS packaging revenue is projected to reach $13.4 billion in 2026, with total back-end revenue hitting $18.3 billion or roungly 10% of revenue. Hyperscaler custom AI-ASICs—Google TPU, AWS Trainium, Meta MTIA, OpenAI’s custom silicon—are projected to consume approximately ~500,000 CoWoS interposer wafers in 2026, generating and via our model rountly $5 billion in packaging revenue. Add front-end die wafer value, and the combined spend reaches approximately ~$9-10 billion annually.
Google TPU alone represents roughly $4 billion in annual wafer and packaging spend. AWS Trainium follows at approximately $2.5 billion. These are volumes that can anchor a meaningful slice of a leading-edge fab module even if Intel only captures a percentage.
The CPU Foundry of Choice
Intel has also historically manufactured, on their process, some of the best CPUs in the market. We think if Intel can become the CPU foundry of choice by aligning its technology roadmap, customer model, and capacity investments with the most structurally advantaged shift in compute demand. Intel doesn’t need a heroic “TSMC replacement” narrative. It needs a credible plan to win a meaningful slice of a CPU foundry market that is already tens of billions of dollars and compounding.
The immediate opening is created by a capacity-and-pricing regime at TSMC that pushes CPU customers to value supply assurance and time-to-market as highly as raw transistor performance. When the market’s dominant supplier is both capacity-tight and pricing-strong, even highly loyal customers become more open to a second trusted foundry—provided it can deliver predictable yields at high clocks and schedule certainty through volume ramp.
The fastest-growing part of CPU demand is shifting toward organizations with both the financial scale and strategic motivation to multi-source. NVDIA and the Hyperscalers are committed to their custom Arm roadmaps. Arm-based processors reached roughly 13% of the general server market by mid-2025, up from 0.5% in 2018. Based on our CPU market model, we believe that primarily thanks to NVIDIA, Arm could reach nearly 30% of the CPU server shipment volume in 2025. It is our conviction that the hyperscalers can and would have structural architectural advantages if they utilized Intel Foundry for their Arm based CPUs.
What Intel Must Deliver
We should point out that all the things we believe Intel Foundry needs to do, Intel Foundry is actually doing. Becoming the foundry of choice in any on vector requires Intel to compete on the customer’s operational metrics, with world class customer support. First, continue to prove out leading-edge CPU manufacturability: fast learning cycles, predictable (and dependable roadmaps), stable PDKs, high parametric yield, and transparent ramp commitments. CPU programs are less forgiving than many ASICs when frequency, leakage, and thermal envelopes are pushed.
Second, Intel must package front-end plus back-end as a single platform. A credible “systems foundry” offer—where customers can secure logic capacity and advanced packaging capacity under one commercial umbrella—directly addresses the pain point emerging as CPUs move to N3/N2 while backend constraints persist.
Third, Intel must win by being the simplest path to production for Arm CPU designers: robust Arm ecosystem enablement, physical IP tuned for high-frequency CPU cores, validated EDA flows, and a customer engagement model that convinces hyperscalers they can protect proprietary microarchitecture while getting best-in-class manufacturing support.
Share Forecast
In a base case where Intel executes competitively and wins one hyperscaler CPU family at meaningful volume plus a portion of the custom ASIC market, Intel could capture 3 to 5% of the hyperscaler Arm CPU and custom accelerator foundry volume by 2028, rising to 8 to 12% by 2030 in several of our scenarios. That translates to several thousand wpm in the late 2020s and potentially (conservatively) 10,000 wpm by 2030 *we have a range of models and are sharing estimates on the conservative side to support the Intel Foundry economics narrative.
Revenue-wise, even a high-single-digit share of a fast-growing Arm CPU subsegment can plausibly translate into low-single-digit billions of annual foundry revenue by the end of the decade—provided Intel meets the operational bar that CPU customers demand. Add some custom ASIC/XPU share and Intel Foundry revenue potential is much higher as outlined in our model below.
What Subscribers Get
The full institutional research report includes:
Complete AI-ASIC data tables — CoWoS interposer wafers, die wafer estimates, packaging revenue, and total spend for Google, AWS, OpenAI, Meta, and others with monthly wafer run-rate calculations
Die wafer economics breakdown — Detailed analysis of the 50/50 split between packaging revenue ($4.7B) and die wafer revenue ($4.6B) for custom AI accelerators
Intel Foundry revenue scenario modeling — Revenue projections at 25%, 50%, and 100% capture rates across all major hyperscaler AI-ASIC programs
TSMC capacity analysis — Node-by-node breakdown of N2/N3/N5 capacity allocation, utilization rates exceeding 100%, and the structural shift that creates Intel’s opening
CPU TAM forecasts — Server shipment projections (17M units in 2026), traditional vs AI server breakdowns, CPU segment 16% CAGR through 2029
Hyperscaler custom CPU roadmap — AWS Graviton, Microsoft Cobalt, Google Axion, and Meta CPU programs with design partners, nodes, volume estimates, and instance growth metrics
Combined opportunity sizing — AI-ASICs plus CPUs at 25% and 50% capture scenarios, establishing the $15-20B addressable range for Intel Foundry
Sensitivity analysis — Low/mid/high case modeling on die wafer conversion factors and pricing assumptions