TSMC U.S. revenue growth hits 75% as Arizona profits reach $1.2B

The concentration of revenue among U.S. firms like Broadcom and Nvidia transforms TSMC from a vendor into critical infrastructure for the American AI and streaming backend. While manufacturing in the Sonoran Desert remains more expensive than in Hsinchu, the $1.2 billion profit suggests that hyperscale cloud providers are willing to pay a premium for supply chain resilience. This shift indicates that localized production is becoming a commercial necessity rather than just a response to the CHIPS Act. For the broader ecosystem, this ensures that the silicon powering next-generation video encoding and AI recommendation engines is less vulnerable to transpacific geopolitical shocks. Watch for TSMC's utilization rates in Arizona as 2nm production approaches to see if domestic margins can stay competitive.

TSMC's Arizona expansion is drawing significant attention from the semiconductor and infrastructure sectors as the company scales domestic production to serve AI-optimized silicon demand. In June 2026, Ericsson launched its AI in RAN software suite that embeds AI models directly into basebands and radios, running on existing AI-ready hardware without requiring new chip deployments. The suite was validated in live commercial trials on T-Mobile's 5G Advanced network, where Ericsson's AI-native scheduler produced roughly 10% better spectral efficiency and up to 15% higher downlink throughput, with full deployment targeted for Q3 2026. These AI-optimized baseband processors represent exactly the class of advanced silicon that TSMC fabricates for customers like Nvidia and Broadcom, underscoring why the foundry's U.S. customer concentration now exceeds 75% of total revenue. The business case for TSMC's domestic manufacturing is being reinforced by the explosive growth in AI-driven network traffic that demands ever more sophisticated silicon. Ericsson's June 2025 Mobility Report found that generative AI traffic represents only 0.06% of total network data traffic currently but is expected to grow substantially as AI agents embed across devices and applications. The report also revealed that AI traffic has a fundamentally different profile from traditional mobile traffic, with 26% uplink versus 74% downlink compared to the typical 10/90 split, creating new demands on network processing hardware. This trajectory toward AI-embedded infrastructure will require increasingly advanced process nodes from TSMC's fabs, and the company's $1.2 billion Arizona profit in H1 2026 demonstrates that domestic production can achieve commercial viability even at higher per-wafer costs than its Taiwan facilities. On the technical side, TSMC's advanced process capabilities are enabling the AI inference hardware that telecom operators are beginning to deploy at scale. Ericsson's AI/ML work in RAN optimization has demonstrated 20-40% CapEx savings in capacity planning and 15% improved spectrum efficiency through pattern recognition, with autonomous energy management reducing radio network consumption by up to 25% without impacting user experience. These AI workloads run on custom silicon requiring leading-edge fabrication, the same advanced nodes TSMC produces in Arizona. The company's domestic capacity ensures that the chips powering next-generation video encoding, , and network optimization functions are less vulnerable to transpacific supply chain disruptions, a consideration that grows more critical as streaming platforms and telecom operators alike embed AI deeper into their infrastructure stacks.