NVDA Plummets -12.8% as Hyperscalers Divert 8.5B to Custom ASICs; Broadcom, Marvell Poised for +15% Upside

Google, AWS, and Meta's Accelerated $85B Custom Silicon Capex Reconfigures AI Chip Landscape, Driving HBM3e Contract Prices Up +22% and Shifting Valuation Multiples Across Semiconductor Sector.

Tradesnaut Quant Research Desk · September 08, 2026 · 6 min read · Semiconductors

NVDA Plummets -12.8% as Hyperscalers Divert 8.5B to Custom ASICs; Broadcom, Marvell Poised for +15% Upside

Key takeaways

Market Dynamics & Earnings Data Breakdown

The semiconductor landscape is undergoing a profound re-rating as major hyperscale cloud providers dramatically accelerate their pivot towards in-house designed ASICs for AI workloads, exerting significant pressure on traditional GPU leaders. Nvidia ($NVDA) shares have experienced a sharp -12.8% decline over the past two weeks, fueled by whispers of Q3 2026 data center revenue growth slowing to an estimated 38% year-over-year, missing previous analyst consensus of 55% and signaling a direct impact from this strategic shift. Sources indicate Google's TPU v6, AWS's Trainium3, and Meta's MTIA v3 now collectively represent between 40-50% of their respective AI compute clusters, a substantial increase from less than 20% just 18 months prior.

Supply Chain Bottlenecks & Macro Valuation Metrics

This hyperscaler pivot has created a nuanced rebalancing within the semiconductor supply chain. TSMC ($TSM), the leading foundry, is a key beneficiary, reporting a +14.5% sequential increase in advanced node (3nm and 2nm) revenue for Q2 2026, largely driven by custom ASIC orders from cloud giants. Its overall foundry revenue share derived from custom ASICs for hyperscalers has climbed to 15% from just 8% in 2024. Despite the moderation in GPU demand, the need for High Bandwidth Memory (HBM) remains insatiable for both GPUs and these new custom ASICs. SK Hynix ($000660.KS) and Samsung Electronics ($005930.KS) have successfully negotiated Q4 2026 contract price increases for HBM3e and HBM4 of +22% and +25% respectively, projecting to expand their memory division operating profit margins by an additional 180-200 basis points. Micron Technology ($MU) is also seeing tailwinds, with its HBM revenue increasing 15% quarter-over-quarter, though still trailing its Korean counterparts in market share. ASML Holding ($ASML) maintains a robust backlog of over €55 billion, demonstrating sustained demand for its leading-edge lithography equipment, essential for these advanced custom silicon designs.

Quantitative Order Flow & Volatility Metrics

Quantitative order flow metrics underscore the market's rapid repricing of the AI chip hierarchy. Over the past month, Nvidia ($NVDA) has witnessed approximately $5.8 billion in institutional net selling, predominantly concentrated in large block trades reflecting a 60% sell-side dominance. The 3-month at-the-money 25-delta put/call skew for NVDA has sharply increased from 0.98 to 1.35, signaling a pronounced bearish sentiment and a demand for downside protection. Concurrently, implied volatility for NVDA 6-month options has surged by 150 basis points to 45%, indicating heightened uncertainty. In stark contrast, Broadcom ($BRCM) and Marvell Technology ($MRVL) have experienced combined institutional net buying of .1 billion, with call option volumes for BRCM running +25% above its 6-month average and its put/call ratio dropping to 0.78 from 1.05. This shift is also palpable in broader market indices; while the SOX Semiconductor Index registered a modest +0.8% gain over the last month, the non-NVDA components within the index gained +3.2%, highlighting the rotation. The KOSPI index, heavily weighted by memory chip giants, posted a +1.8% gain, bolstered by the strong HBM pricing environment and positive outlook for SK Hynix and Samsung Electronics.

Tags: Memory Chips, SK Hynix, Semiconductors, Wall Street, AI Chips, ASICs, Hyperscalers