Daily Top-10 Movers

Prices are Wednesday, September 23, 2026 closes at 4:00 PM EDT. Ranked by absolute price change among semiconductor stocks with market cap ≥$5B.

★ highlighted = today's news-driven movers.

Broadcom Rides Google's TPU Ramp as Intel's Foundry Losses Cloud Its Turnaround

• Google's TPU v8, manufactured with Broadcom, delivers roughly three times the training performance per watt of its predecessor and about 1.8 times the inference performance, with Broadcom shipping the inference-focused TPU 8i in the third quarter and volume ramping in the fourth, per Seoul Economic Daily; Morgan Stanley projects direct TPU sales of $84 billion in 2027 and $108 billion in 2028 as Google supplies external data centers and Anthropic scales its TPU footprint from 1 gigawatt this year to 5 gigawatts in 2027. Why it matters: a Morgan Stanley model putting standalone TPU revenue in the same range as merchant GPU franchises gives Broadcom's custom-silicon business an independently sized market of its own — and adds to the case pressuring Nvidia's valuation multiple, which closed at a decade low earlier this week.
• Intel Foundry Services posted a $2.1 billion operating loss in the second quarter even as total company revenue grew 25.4% year over year and Data Center and AI revenue surged 59%, per 24/7 Wall Street, which pegs Wall Street's consensus price target at $116.37 against Wednesday's $122.60 close. Why it matters: Intel shares have more than tripled over the past year on turnaround optimism, but the foundry division's continued losses remain the single biggest swing factor in whether that re-rating holds — an external customer win would do more for the multiple than further data-center growth on the product side.

Samsung and SK Hynix Eye a Combined $139 Billion Quarter as DRAM Outvalues TSMC's Wafers

• Samsung Electronics and SK Hynix are tracking toward a combined 189.9 trillion won (about $139 billion) operating profit in the third quarter, split roughly 111.8 trillion won for Samsung and 78.1 trillion won for SK Hynix, per BigGo Finance, citing FnGuide consensus estimates; server DRAM prices are projected to rise another 13-18% and NAND flash 10-15% in the quarter, per TrendForce data cited in the same report, while Samsung's HBM4 revenue is projected to more than triple quarter over quarter. Why it matters: SK Hynix's implied quarterly profit alone would exceed its combined 2024-2025 operating profit, underscoring how completely HBM and DRAM pricing now drive memory-maker earnings — and how far Q4 guidance from both companies, and from Micron on September 30, would have to fall to disappoint.
• The value of a square millimeter of the newest 1b DRAM has climbed to roughly $0.654, about 50% above the per-area value of TSMC's 2-nanometer wafer at $0.424 and above 3-nanometer's $0.283, with DDR5 16-gigabit chips now averaging $24.80, per Sammy Fans, citing ZDNet Korea. Why it matters: when memory silicon is worth more per square millimeter than logic on the industry's most advanced node, it helps explain why Samsung and SK Hynix are posting record profits even without matching Nvidia's or TSMC's headline growth rates — and why capital keeps shifting toward memory fabs at the margin.

AT&S Expands Its Marvell Substrate Deal as TSMC's Price Hike Pulls Forward Equipment Orders

• AT&S and Marvell Technology signed an expanded long-term agreement for advanced IC substrates supporting AI accelerators and the shift from monolithic chips to chiplet-based designs, per Evertiq, with AT&S fitting out Plant 2 and building a new facility for IC substrate cores at its Kulim, Malaysia site. Why it matters: substrate capacity, not wafer starts, has been one of the quieter bottlenecks in AI-chip supply chains — a multi-year commitment from a customer like Marvell gives AT&S the revenue visibility to justify the capex, and signals packaging supply is becoming as contested as leading-edge silicon itself.
• TSMC's 45-nanometer-and-below capacity is running fully loaded and mature and specialty lines are tightening too, extending customer order visibility out to 2030, per GuruFocus, beyond the headline wafer-price increase TSMC plans for January 2027. Why it matters: a shortage spanning leading-edge and legacy nodes alike pulls forward equipment orders industry-wide — not just at ASML and Lam Research for 2-nanometer tools, but for the trailing-edge and specialty toolmakers serving TSMC's older, auto- and industrial-facing fabs too.

Huawei and Alibaba Give Xi a Chip Card as Export Controls Stay Off the Summit Agenda

• Huawei accelerated its next AI chip's release by three quarters and unveiled a new SuperPoD architecture linking fewer processors than originally planned, while Alibaba unveiled its Zhenwu M890 chip and DeepSeek assembled the largest known cluster of Huawei chips to date, per CNBC, timed just ahead of Xi Jinping's state visit to Washington. Asia Society Policy Institute fellow Lizzi Lee said the message to Trump is that "you can't simply technologically choke us off," while Tufts University's Chris Miller noted U.S. AI firms still earn 50 to 100 times the revenue of their Chinese counterparts. Why it matters: the timing hands Xi negotiating leverage on export controls without changing the underlying revenue math — Chinese chipmakers are narrowing the technology gap on paper, but the near-term addressable market for Nvidia and AMD in China is unaffected as long as the revenue gap Miller describes persists.
• Nvidia remains barred from selling its most advanced accelerators to China even as Trump has allowed H200 chip sales to "approved customers," and Washington is separately seeking greater access to Chinese rare earths and critical minerals at Thursday's summit, per CBS News and Outlook India, with Beijing continuing to treat those mineral export controls as negotiating leverage. Why it matters: the H200 status quo keeps Nvidia's constrained-but-real China channel open through the summit, while a rare-earths concession would matter more to equipment and packaging suppliers than to chip designers — gallium, germanium and related inputs feed etching and packaging steps across the value chain, not just one company's bill of materials.

CLSA Sees a Chip Shortage Into 2030 as Wolfe Lifts SK Hynix on Free Cash Flow

• CLSA's Bhavtosh Vajpayee estimates demand for GPUs and custom AI chips currently exceeds supply by roughly 73% and expects the shortage to persist until around 2030, per CNBC (published September 23), calling talk of an AI-semiconductor demand peak in 2028 "utterly wrong."
• Wolfe Research's Chris Caso raised his SK Hynix price target to $250 from $200, reiterating an Outperform rating, per TradingView (published September 23), writing that "while much good news is known, we think the cash flow will be difficult to ignore" as SK Hynix and Micron are projected to generate enough free cash flow in 2026-2027 to fund buybacks worth 32% and 25% of their market caps, respectively.

OpenAI's DevDay and Micron's Earnings Bracket Next Week

• Tuesday, September 29 — OpenAI holds its annual DevDay in San Francisco, the company's largest developer event of the year, per OpenAI.
• Wednesday, September 30 — Micron Technology reports fiscal fourth-quarter results, with the HBM4 ramp and gross-margin guidance in focus, per Investing.com.