Samsung's 2nm Gamble, TSMC Revenue Surge & Intel's High-NA Leap

Jamie Call (0:00)

A Hardware & Chips Daily News.

I'm Jamie Call, thanks for joining me.

Today, the Dual Prime Core Gamble.

Samsung has started mass production on the Exynos 2700, its second-generation 2nm chip, and the architecture choice at the center of it is worth paying close attention to.

The Exynos 2700 runs on Samsung's SF2P process, an enhanced version of the node used in the Exynos 2600

It uses Gate-All-Around transistors, the same fundamental approach as its predecessor, but this time Samsung has committed to a Dual Prime Core design, targeting 4.24 GHz.

The initial production batch is 10% larger than the previous generation, and early Geekbench 6 multicore numbers show a 9.5% lead over Qualcomm's Snapdragon 8 Elite Extreme Gen 6

Here's what matters about that figure. It's a performance number only. There are no efficiency metrics, no power draw data, no thermal benchmarks attached to it. A 9.5% multicore lead could easily be erased if that frequency target comes with meaningful battery drain or sustained throttling. The Dual Prime Core bit signals Samsung's intent to compete on peak performance. Whether it can do so without a thermal penalty is the unresolved question that will determine how much of the Galaxy S27 lineup actually ships with this chip. The architectural choice here is more than a spec sheet decision. Qualcomm runs a single prime core strategy. Samsung is running too. The implication is a deliberate diversions in how each company thinks flagship workloads are evolving. Sustained multi-threaded tasks, AI inference at the edge, and concurrent process handling all favor dual-prime design on paper. The trade-off is silicon area and power budget management. Samsung is making a calculated bet that the workload nicks on next-generation devices justifies that cost. The benchmark lead is real, but until thermal and efficiency data surfaces, treating it as a confirmed advantage would be premature. While Samsung is ramping in Korea, TSMC is accelerating on its own two nanometer front. The N2 node has reached 3% of TSMC's second quarter revenue, which came in at $40.2 billion.

High-performance computing now represents 66% of that revenue, with 20% quarter-on-quarter growth. Management has raised full-year 2026 guidance to slightly above 40% growth in US dollar terms, up from a prior expectation closer to 30%. The signal here is clear. AI hardware demand is pulling TSMC's advanced node ramp faster than the company itself guided for six months ago. The next-generation A14 node is already planned for 2027, with a 10% to 15% speed improvement over N2. TSMC is building runway, not just filling current orders. Here's the thing. The margin picture is more complicated. 2 nanometer production carries 3% to 4% points of gross margin compression versus older nodes. A $60 billion plus capital expenditure envelope raises legitimate questions about profitability sustainability.

So, advanced packaging scarcity is providing some pricing power offset. But the economics at the leading edge are tighter than the revenue headline suggests. There's a newer and less certain development worth flagging. TSMC is exploring a potential partnership with Elon Musk's Terafab project, a proposed $119 billion ship manufacturing facility. The roles on the discussion include an equity stake, operational control or technology and capital provision. Intel is also reportedly involved in early conversations around the Texas-based facility. The key implication is strategic hedging.

TSMC derives roughly 75% of wafer revenue from North American customers. A US domestic presence reduces concentration risk tied to Taiwan's strait tensions. But Terafab's financial plan has a credibility problem. Analyst estimates put the capital required for Terafab's stated capacity goals somewhere between $5 and $13 trillion. The $119 billion figure doesn't close that gap. TSMC's willingness to engage depends heavily on actual customer demand signals, Musk's financial stability and a realistic construction timeline. These are early discussions, not a commitment. Intel's separate story this cycle is about manufacturing credibility. The company has moved high-end EUV lithography beyond qualification and into production on selected Panther Lake processor layers, in collaboration with ASML. That's a meaningful step. Moving from qualification to production means Intel is generating real yield data on the world's most advanced lithography tool. Consider this. The implication for Intel's foundry ambitions is straightforward to state and hard to deliver. Technology milestones matter, but the stock trades at a significant premium to its assessed fundamental value. What justifies that premium is paying customers at volume scale, not production process announcements. Intel has the tool. The question the market is asking is who will pay to use it. Pulling back. The through line across all of this is a 2nm race running on two tracks simultaneously. Samsung is betting on dual-prime performance. TSMC is betting on node maturity, packaging scarcity and US geographic diversification. Intel is betting it can convert lithography leadership into foundry revenue before its balance sheet runs out of patience. The near-term signals worth watching are Samsung's thermal and efficiency disclosures for the Exynos 2700, TSMC's gross margin trajectory is N2 scales further, and whether Terafab produces any credible customer or financing announcements. Those three data points will tell us considerably more than the headline numbers already have. Thanks for listening. This podcast was built using AI technology. A Yes We production.