AMD Hits Record x86 Share While TSMC Closes A16 Development and Outsources Packaging to Intel

AMD's x86 market share hit an all-time high of 34.1 percent in the second quarter of 2026, up 4.7 percentage points year over year, according to fresh Mercury Research data published on August 21. The disclosure arrived on the same day Taiwan's TSMC confirmed it had completed development and verification of its 1.6-nanometer A16 process — the industry's first commercial backside-power-delivery node — and revealed it has begun outsourcing part of its CoWoS advanced packaging work to Intel's Malaysian plant. Together, the three announcements reset the competitive map for AI and PC silicon heading into the back half of 2026.

AMD's Biggest Quarter in x86 History

AMD's client x86 share crossed the 30 percent threshold for the first time in the company's history, reaching 30.3 percent of desktop-and-mobile combined shipments in Q2 2026. The desktop share rose to 34.9 percent (up 1.8 points sequentially) and the mobile share hit 28.9 percent (up 0.6 points). In servers, AMD's share climbed to 34.5 percent excluding IoT — a 7.3-point year-over-year gain. Mercury Research president Dean McCarron attributed the gains to higher demand for EPYC server CPUs and Ryzen AI PC parts, and noted that AMD's revenue per client unit hit a fresh record even as desktop shipments across the industry fell more than 20 percent year over year on price pressure and limited GPU supplies.

TSMC's A16 Hits the Tape

TSMC confirmed on August 20 that A16, the foundry's first 1.6-nanometer process and the industry's first commercial backside-power-delivery (BSPDN) node, has completed development and verification. Mass production is scheduled for the fourth quarter of 2026, with first customer silicon expected in 2027 and systems in 2028. Industry coverage this week confirmed Nvidia has booked A16 capacity for its Feynman GPU architecture, which follows the Rubin family and is slated for production in the second half of 2028. TSMC also reaffirmed that A14 remains on track for 2028 mass production, with A13 and A12 scheduled for 2029. TSMC's board approved a $29.4 billion capital expenditure plan for advanced processes and packaging earlier this month.

Intel Picks Up CoWoS Overflow

In a separate disclosure on August 18, TSMC acknowledged that it has begun outsourcing part of its back-end CoWoS work to Intel's Malaysia plant. The shift showed up in export data: roughly $1.3 billion of high-bandwidth memory originally slated for Taiwan was redirected to Malaysia, while HBM volumes entering Taiwan fell sharply to under $3 billion. Intel's local plant is the only facility in Malaysia with the scale and capability to assemble HBM into AI accelerator products. CoWoS lead times currently run 52 to 78 weeks, meaning an order placed today would not deliver until after mid-2028. Intel's own advanced-packaging technology, EMIB-T, is expected to reach mass production next year and is priced at roughly half of CoWoS.

Why This Matters for the AI Compute Stack

The three announcements together mark the most significant structural shift in the semiconductor industry since the launch of 3-nanometer production. AMD's continued share gain at Intel's expense narrows the historical x86 duopoly to its tightest level in modern times and gives AMD's data-center unit the volume base needed to fund aggressive MI400 and MI500 accelerator roadmaps. TSMC's A16 completion, with backside power delivery as the headline feature, locks in a process lead over Samsung — which pushed its 1.4-nanometer mass production target to 2029 at the SAFE Forum in February — and reasserts TSMC's pricing power for AI accelerators through 2027. Intel's emergence as a CoWoS subcontractor, finally, gives the broader AI accelerator ecosystem a credible second source for advanced packaging, which has been the single largest bottleneck for years.

What to Watch Through Year-End

Three checkpoints follow. TSMC's September monthly revenue release will be the first to include A16 pre-production wafer revenue. AMD's Q3 earnings on or around October 28 will show whether the Q2 share gains are translating into ASP expansion on EPYC and Instinct. And Intel's third-quarter earnings will reveal how much of the CoWoS overflow has been booked, and whether Intel's foundry services segment — long a drag on margins — can finally contribute positive operating income.