top of page

November 21, 2026

Nvidia Slump Deepens as Power Limits and Margin Scrutiny Hit AI

NVDA, AMD, AVGO, AAPL, META, GOOGL, TTWO, ZS, CRWD, PANW

Executive Summary

  • AI Sell-Off & IPO Reality: Tech stocks slid sharply as Anthropic's historic $119.5 billion IPO faced second-day volatility amid mounting AI cost and margin scrutiny.

  • Chip & Power Bottlenecks: Semiconductors tumbled on investigations, short-seller claims, and power limits that slowed data center builds and custom chip orders.

  • Agent & Cyber Shifts: Companies are pivoting toward AI agents and cybersecurity consolidation even as model prices drop and software faces valuation pressure.

  • Gaming Blockbuster Divergence: Grand Theft Auto 6 hit record pre-orders and reviews while Take-Two shares dropped on high pre-launch expectations.


Corporate Dynamics

Anthropic: Historic $120 Billion Offering Meets Second-Day Volatility and Margin Scrutiny

  • News: Shares of Anthropic—the first publicly traded frontier AI lab—fell roughly 10% to 11% to trade between $85.92 and $87 in its first full trading session after surging as much as 39% and closing up around 17% on debut (where shares reached nearly $112). Anthropic exercised its full green shoe option of 147 million shares at the $83 IPO price to reach a $119.5 billion total offering size (referenced by management as $120 billion), while reporting $81 billion in annual recurring revenue (ARR) per Bloomberg Intelligence's Mandeep Singh, $54 billion in revenue against roughly $900 billion in long-term computing commitments, and a drop in compute cost per dollar of revenue from $2.51 down to $0.60 per T. Rowe Price's Emma Norchet (also referenced as Emma Nerchet). Per fund manager Dan Niles, combined sequential revenue growth for Anthropic and OpenAI slowed from 75% in Q2 to 48% in Q3 while combined operating losses rose from $7.9 billion to $8.4 billion amid price competition, and Anthropic's margins contracted from Q2 to Q3. Chief Financial Officer Krishna Rao noted the capital supports execution across multi-platform compute and over 20 years of supplier commitments, CEO Dario Amodei told Jamie Dimon the lab buys less compute upfront and uses last-minute spot markets despite acknowledging in its IPO filing that underestimating computing needs set back products this year, board member and Novartis CEO Vas Narasimhan advocated for appropriate AI regulation, and early backers including MGX (holding an unrealized gain of roughly $20 billion), T. Rowe Price, Menlo, and GIC participated alongside a low-key San Francisco culture.

  • Outlook: Lower compute cost per revenue dollar and high-spending enterprise customers—with the top 250 each spending over $10 million while holding just 0.1% of Fortune 500 budgets—support long-term expansion into coding agents, knowledge work, security, biology, and robotics. At the same time, margin compression, slower sequential growth across frontier AI labs, and heavy compute commitments create near-term pressure on profitability and execution.


NVDA (Nvidia): Chip Slide Deepens Amid Investigation, Leadership Fallout, and Rubin Financing

  • News: Nvidia shares fell 6% in one session—marking a seventh straight decline, its longest losing streak since August, and hitting a low since May 2025 following the resignation and downfall of Jensen Huang—and later dropped by double digits to bring its five-day loss since Monday to one-fifth of its value. The company faces a California investigation into whether its actions allowed sensitive AI chips and confidential information to reach China improperly, and newly released congressional committee documents showed the Trump administration approved nearly 900,000 Nvidia and AMD chips for China between December and May, including to firms later added to a Pentagon list for aiding China's military, though Nvidia stated those sales carried strict conditions and volume limits. Meanwhile, SpaceX's $32 billion debt deal will fund a trust to buy Nvidia's new Rubin chips for roughly five-year leasing, and Anthropic CFO Krishna Rao reaffirmed Nvidia as a critical investor, strategic partner, and technology supplier.

  • Outlook: Demand from AI hardware financing like SpaceX's Rubin chip trust and ongoing compute partnerships supports chip shipments, yet regulatory investigations, export-control scrutiny, and leadership transition weigh heavily on investor confidence and near-term sentiment.

Get instant access to AI-driven US tech market research.

Executive Summary

  • AI Sell-Off & IPO Reality: Tech stocks slid sharply as Anthropic's historic $119.5 billion IPO faced second-day volatility amid mounting AI cost and margin scrutiny.

  • Chip & Power Bottlenecks: Semiconductors tumbled on investigations, short-seller claims, and power limits that slowed data center builds and custom chip orders.

  • Agent & Cyber Shifts: Companies are pivoting toward AI agents and cybersecurity consolidation even as model prices drop and software faces valuation pressure.

  • Gaming Blockbuster Divergence: Grand Theft Auto 6 hit record pre-orders and reviews while Take-Two shares dropped on high pre-launch expectations.


Corporate Dynamics

Anthropic: Historic $120 Billion Offering Meets Second-Day Volatility and Margin Scrutiny

  • News: Shares of Anthropic—the first publicly traded frontier AI lab—fell roughly 10% to 11% to trade between $85.92 and $87 in its first full trading session after surging as much as 39% and closing up around 17% on debut (where shares reached nearly $112). Anthropic exercised its full green shoe option of 147 million shares at the $83 IPO price to reach a $119.5 billion total offering size (referenced by management as $120 billion), while reporting $81 billion in annual recurring revenue (ARR) per Bloomberg Intelligence's Mandeep Singh, $54 billion in revenue against roughly $900 billion in long-term computing commitments, and a drop in compute cost per dollar of revenue from $2.51 down to $0.60 per T. Rowe Price's Emma Norchet (also referenced as Emma Nerchet). Per fund manager Dan Niles, combined sequential revenue growth for Anthropic and OpenAI slowed from 75% in Q2 to 48% in Q3 while combined operating losses rose from $7.9 billion to $8.4 billion amid price competition, and Anthropic's margins contracted from Q2 to Q3. Chief Financial Officer Krishna Rao noted the capital supports execution across multi-platform compute and over 20 years of supplier commitments, CEO Dario Amodei told Jamie Dimon the lab buys less compute upfront and uses last-minute spot markets despite acknowledging in its IPO filing that underestimating computing needs set back products this year, board member and Novartis CEO Vas Narasimhan advocated for appropriate AI regulation, and early backers including MGX (holding an unrealized gain of roughly $20 billion), T. Rowe Price, Menlo, and GIC participated alongside a low-key San Francisco culture.

  • Outlook: Lower compute cost per revenue dollar and high-spending enterprise customers—with the top 250 each spending over $10 million while holding just 0.1% of Fortune 500 budgets—support long-term expansion into coding agents, knowledge work, security, biology, and robotics. At the same time, margin compression, slower sequential growth across frontier AI labs, and heavy compute commitments create near-term pressure on profitability and execution.


NVDA (Nvidia): Chip Slide Deepens Amid Investigation, Leadership Fallout, and Rubin Financing

  • News: Nvidia shares fell 6% in one session—marking a seventh straight decline, its longest losing streak since August, and hitting a low since May 2025 following the resignation and downfall of Jensen Huang—and later dropped by double digits to bring its five-day loss since Monday to one-fifth of its value. The company faces a California investigation into whether its actions allowed sensitive AI chips and confidential information to reach China improperly, and newly released congressional committee documents showed the Trump administration approved nearly 900,000 Nvidia and AMD chips for China between December and May, including to firms later added to a Pentagon list for aiding China's military, though Nvidia stated those sales carried strict conditions and volume limits. Meanwhile, SpaceX's $32 billion debt deal will fund a trust to buy Nvidia's new Rubin chips for roughly five-year leasing, and Anthropic CFO Krishna Rao reaffirmed Nvidia as a critical investor, strategic partner, and technology supplier.

  • Outlook: Demand from AI hardware financing like SpaceX's Rubin chip trust and ongoing compute partnerships supports chip shipments, yet regulatory investigations, export-control scrutiny, and leadership transition weigh heavily on investor confidence and near-term sentiment.


AMD: Short-Seller Allegations and Export Scrutiny Pressure Shares

  • News: AMD shares dropped another 9% on Friday, putting the company on track for its steepest weekly decline since November 2025, after short-seller Hunterbrook Capital (also referenced as Hunter Brook Capital) took a short position and its media affiliate alleged risky business ties in China. Newly released congressional committee documents also covered nearly 900,000 Nvidia and AMD chips approved for sale to China between December and May, including to entities later placed on a Pentagon list. AMD denied wrongdoing, stating the report misrepresents its practices and that it maintains strict compliance with U.S. export control laws and regulations.

  • Outlook: Aggressive refutations and export-control compliance aim to protect its commercial standing, but short-seller scrutiny and U.S.-China chip oversight keep regulatory and reputational risk elevated.


AVGO (Broadcom): Custom Chip Order Cut Rumor and Power Limits Hit Shares

  • News: Broadcom shares dropped more than 10% amid a broad semiconductor sell-off and an unconfirmed report from analyst Ming-Chi Kuo that Google cut roughly 20% of its custom AI chip orders—designed with Broadcom—from next year's plans, shifting some orders to rival MediaTek. Kuo attributed the cut to electricity shortages constraining Google's data center capacity.

  • Outlook: Physical power bottlenecks at data centers can directly trim custom silicon orders even when underlying chip demand remains high, creating near-term order risk for custom chip design partners.


MediaTek: Potential Beneficiary of Custom Chip Order Shifts

  • News: According to an unconfirmed report from analyst Ming-Chi Kuo, Google is shifting some custom AI chip orders—originally designed with Broadcom—toward MediaTek after trimming roughly 20% of next year's planned orders due to data center electricity constraints.

  • Outlook: Picking up reallocated custom AI chip orders offers a potential volume offset, though overall order growth remains tied to data center power availability.


OpenAI: Rapid Usage Growth Contrasts With Monetization and Margin Pressures

  • News: Per fund manager Dan Niles, combined sequential revenue growth for OpenAI and Anthropic slowed from 75% in Q2 to 48% in Q3 while combined operating losses widened from $7.9 billion to $8.4 billion amid price competition, and Niles noted only 10% of OpenAI's consumer-heavy installed base pays for services. OpenAI is also an early lock-in buyer of large-scale compute capacity and faces competition in consumer agents from Apple, Meta's Muse and Instinct, and Anthropic, while Abu Dhabi's MGX holds stakes across OpenAI, Anthropic, and xAI.

  • Outlook: Converting a small paying share of a large consumer base while competing against falling model prices and rival agent ecosystems keeps monetization and operating losses central to its business model.


xAI: Included Among Major AI Labs Backed by MGX

  • News: Abu Dhabi's MGX holds a stake in xAI alongside investments in Anthropic and OpenAI.

  • Outlook: Backing from a major sovereign-linked investor provides capital access in a capital-intensive AI race, though competition across frontier labs remains fierce.


AAPL (Apple): Developing Device-Level AI Agents Amid Hardware and Model Questions

  • News: Citing job listings reported by Mark Gurman, Bloomberg reported that Apple is building an AI agent capable of writing code, controlling apps, and navigating devices on a user's behalf. Bloomberg Intelligence's Mandeep Singh noted Apple faces tough competition in consumer agents from Meta's Muse and Instinct, OpenAI, and Anthropic and questioned whether Apple can match Anthropic's Claude precision given limited prior proprietary model investment, while Factory co-founder and CTO Eno Reyes noted Apple could use Mac hardware if it lifts its current two-virtual-machine limit per Mac to run CPU-bound personal agents locally once open-weight models like Google's Gemma run efficiently standalone on Apple chips.

  • Outlook: Tight hardware-software integration and on-device execution offer a distinct consumer angle, but catching up on model precision and adjusting local virtualization limits will shape how quickly its agent tools gain traction.


META (Meta): Competing in Consumer Agents With Muse and Instinct

  • News: Bloomberg Intelligence's Mandeep Singh identified Meta's Muse and Instinct among the consumer AI agent rivals competing against Apple's planned agent, OpenAI, and Anthropic.

  • Outlook: Pushing Muse and Instinct positions Meta in the consumer assistant race, where user habit and model capability will determine adoption against established AI platforms.


GOOGL (Google): Lower Model Pricing and Power-Constrained Data Center Adjustments

  • News: Data from developer service OpenRouter shows model prices across Google, Anthropic, and OpenAI fell 57% since January while usage jumped more than 13-fold and total spending rose roughly six-fold, though total spending flattened over the past five weeks as users shifted via auto-model routing toward cheaper and open-weight models. Factory co-founder and CTO Eno Reyes cited Google's open-weight Gemma models as an example of models that could eventually run standalone on Apple chips, while an unconfirmed report from analyst Ming-Chi Kuo stated Google cut about 20% of next year's custom AI chip orders designed with Broadcom due to electricity limits on data center capacity.

  • Outlook: Falling model prices and open-weight options expand developer usage while pressuring per-query pricing, and power limits on data centers can force adjustments to hardware buildouts.


SpaceX: Structuring $32 Billion Debt Deal to Lease Nvidia Rubin Chips

  • News: SpaceX arranged a $32 billion debt deal to fund a trust that will purchase Nvidia's new Rubin chips and lease them to SpaceX over roughly five years, with most of the debt priced at yields above 8%.

  • Outlook: Leasing high-end AI hardware through a dedicated trust lets SpaceX access Rubin chips without buying them outright on its own balance sheet, though yields above 8% show lenders require a notable return for AI infrastructure risk.


TTWO (Take-Two Interactive): Grand Theft Auto 6 Breaks Records While Stock Pulls Back

  • News: Sensor Tower data showed Grand Theft Auto 6 generated 24.5 million pre-orders—roughly $2.3 billion in pre-order revenue—and earned a record 99 out of 100 on Metacritic (beating Red Dead Redemption 2 at 97 and the original 1998 N64 The Legend of Zelda: Ocarina of Time), with Bloomberg's Jason Schreier calling the Vice City title a "masterpiece" after more than 45 hours of play. Despite this, Take-Two shares suffered their worst week since September of this year—and at one point their worst week since November 2022—leaving the stock down 19% year-to-date, which Wedbush Senior VP of Equity Research Alicia Reese (who maintains an Outperform rating and a $280 price target) attributed to "sell-the-news" trading after unit expectations reached over 40 million while Sensor Tower data implied a $95 to $96 average selling price skewed toward the $100 Ultimate Edition over the $80 Standard tier.

  • Outlook: Strong pre-order revenue, higher-priced Ultimate Edition demand, upcoming PC releases, and high-margin recurring GTA Online revenue support multi-year earnings potential even when initial launch hype cools.


Ubisoft: Forecasts Holiday Bookings Drop Amid Crowded Slate and GTA 6

  • News: Ubisoft warned of a tough holiday season, projecting a 26% year-over-year decline in bookings amid a packed release schedule dominated by Grand Theft Auto 6, leaving its shares down more than 40% for the year.

  • Outlook: Competing directly against a blockbuster release like GTA 6 during the holiday window pressures bookings and market share, making rival timing and franchise differentiation critical.


Carta: Shifting Engineering Toward Model Context Protocol and Cloud Desktops

  • News: CEO Henry Ward stated Carta has slowed development on Carta.com to build Model Context Protocol (MCP) apps inside "cloud desktop" and "codeex" architectures, treating Anthropic's platform as a new operating system rather than just an app. Ward also noted that while Carta customers raised roughly $50 billion from venture capital in a single quarter last year, early-stage fundraising is tightening—pointing to a startup growing from $2 million to $20 million questioning if its metrics fit a Series B—and said $10 billion to $20 billion private software companies face a "SaaS apocalypse" narrative in public markets.

  • Outlook: Building directly into emerging AI agent and MCP environments aims to keep Carta embedded in modern workflows, while tighter venture funding and public software skepticism raise the bar for startup fundraising and exits.


Factory: Seeing Auto-Model Routing and Expanding Into Cybersecurity

  • News: Co-founder and CTO Eno Reyes noted that over the last five weeks total developer spending flattened on OpenRouter despite falling model prices and higher usage of cheaper, open-weight models because customers use auto-model routing to send simpler tasks to cheaper models, while U.S. policy changes and national security concerns add uncertainty around choosing open models. Reyes also noted Factory is expanding into cybersecurity offerings.

  • Outlook: Smart model routing helps customers control AI bills and pushes vendors to offer cost-efficient orchestration and adjacent security tools.


ZS (Zscaler): Shares Jump on Report of CrowdStrike Acquisition Talks

  • News: Zscaler shares rose sharply in extended trading after a Bloomberg report said CrowdStrike is in talks to acquire the company, in a deal that would form an entity valued at roughly $265 billion.

  • Outlook: Potential combination talks highlight ongoing consolidation among cybersecurity platforms seeking broader enterprise security coverage.


CRWD (CrowdStrike): Reported in Talks to Acquire Zscaler

  • News: According to a Bloomberg report, CrowdStrike is in talks to acquire Zscaler, which would create a combined business with an estimated market capitalization of around $265 billion.

  • Outlook: Acquiring Zscaler would scale CrowdStrike's footprint across network and cloud security, though deal terms and regulatory review remain key factors to watch.


PANW (Palo Alto Networks): Acquires Applied Compute to Integrate Specialized AI Models

  • News: Factory co-founder and CTO Eno Reyes noted Palo Alto Networks acquired Applied Compute to integrate post-training and cyber-specific models.

  • Outlook: Bringing specialized AI model capabilities in-house supports faster threat detection and security automation as attacks grow more automated.


Databricks: Expanding Into Cybersecurity Offerings

  • News: Factory co-founder and CTO Eno Reyes noted Databricks is expanding into cybersecurity offerings alongside Palo Alto Networks and Factory.

  • Outlook: Extending data platforms into security lets companies analyze telemetry and enforce protections closer to where enterprise data lives.


Mercedes-AMG Petronas F1 Team: Pairs High-Speed AI Simulations With Driver Judgment

  • News: At Microsoft's Ignite conference, Team Principal Toto Wolff and Microsoft Commercial Business head Judson Althoff discussed how the 2,500-person operation uses AI alongside computational fluid dynamics (CFD) and wind tunnels to run over 10 million race-day simulations for millisecond-level strategy choices, while Wolff noted human driver instinct still overrides data when the two conflict and Althoff stressed strict governance, observability, and security for autonomous agents so patch speed outpaces cyberattack speed.

  • Outlook: Fast simulation and agent tooling can sharpen real-time operational decisions, while human judgment and security controls remain essential guardrails.


Industry Trends

Data Center Power Limits and Hardware Financing Shifts

  • Analysis: Even though overall chip sales remain near record levels—nearly two and a half times higher than a year ago—electricity constraints and local pushback are slowing physical buildouts. Forty out of 41 recent midterm ballot initiatives to halt data center construction passed, governors flipped in Texas, Georgia, Ohio, New Mexico, and Michigan on platforms aimed at halting or slowing data center builds, and power limits were cited behind Google's reported trim to custom AI chip orders. At the same time, investor and early Anthropic backer Mida called the market a "GPU wastage bubble" due to spiky research compute needs, while SpaceX's $32 billion debt-funded chip-leasing trust shows companies turning to specialized financing as borrowing costs stay elevated.

  • Outlook: Chipmakers, utilities, and AI labs must navigate grid bottlenecks, local permitting resistance, and higher financing costs, shifting advantage toward operators that match compute capacity to real workload demand.


Falling Model Prices, Auto-Routing, and Platform Realignment

  • Analysis: OpenRouter data shows model prices across Anthropic, OpenAI, and Google dropped 57% since January while usage rose more than 13-fold and spending rose roughly six-fold, before flattening over five weeks as auto-model routing steered simpler tasks to cheaper or open-weight models. Simultaneously, enterprise software is adjusting as Carta builds MCP apps inside cloud desktop and codeex environments, startups face tighter early-stage funding after last year's $50 billion quarterly VC pace on Carta, $10 billion to $20 billion private software firms confront a "SaaS apocalypse" narrative, and cybersecurity consolidates around platforms and specialized AI models.

  • Outlook: Cheaper models and agent-native workflows lower unit intelligence costs for buyers while forcing software and security vendors to prove durable value and protect against faster cyber threats.


Geopolitics and U.S. AI Policy Friction

  • Analysis: U.S.-China tech friction remains a central overhang, marked by newly released congressional disclosures on nearly 900,000 Nvidia and AMD chips approved for China between December and May, new restrictions blocking Chinese firms from accessing advanced AI chips via third-country cloud services, and President Trump's trip to Southeast Asia and Shenzhen—including bilateral talks with Chinese President Xi Jinping and a trilateral meeting with Xi and Russian President Vladimir Putin. U.S. Trade Representative Jamieson Greer pointed to "robust discussion" balancing trade stability and national security, while Harvard Belfer Center's Julian Gewirtz criticized limited transparency and enforcement details and noted figures like Elon Musk and David Sacks alongside Scott Bessent, Marco Rubio, and Howard Lutnick vying for influence over AI policy.

  • Outlook: Evolving export rules, cloud access limits, and shifting policy voices create ongoing compliance and supply-chain uncertainty for hardware makers and model developers operating globally.


Market Sentiment

Cautious Rotation Out of High-Multiple Tech Into Safety and Proof of ROI

  • Analysis: Price action shows broad de-risking across tech: the S&P 500 entered correction territory (down 10% from its October record) on a seven-session losing streak—its longest since February 2020—the Nasdaq dropped over 2% in one session and later fell more than 5% on Friday, November 20 to turn down roughly 3% year-to-date, the Nasdaq 100 hit its lowest level since April, the Magnificent 7 index fell 8% year-to-date, the Philadelphia Semiconductor Index (SOX) slid nearly 7%, and investors moved from equities into bonds from London to Hong Kong to New York. This aligns with Section 1 and Section 2 data showing AI lab operating losses rising to $8.4 billion alongside slower combined sequential revenue growth, model prices falling 57%, and power bottlenecks or investigations hitting chipmakers.

  • Outlook: Markets are demanding clearer returns on massive AI spending rather than paying peak multiples on potential alone, keeping volatility high around earnings, margins, and infrastructure bottlenecks.


Debate Over AI Enablers vs. Adopters and Valuation Disconnects

  • Analysis: AB (AllianceBernstein)'s Leu (also referenced as Lechu) sees AI adoption nearing a 2027 inflection point as leadership shifts from pure "AI enablers" like chip and power providers toward rapid adopters across healthcare, financial services, and technology over five- to ten-year horizons, while Henry Ward highlights a gap between protected private startup valuations and volatile public common stock. Views also diverge on execution—such as high-yield debt demand (above 8%) on SpaceX's chip-leasing deal versus strong enterprise coding-agent demand for Anthropic and record GTA 6 pre-orders colliding with "sell-the-news" trading in Take-Two.

  • Outlook: Expect sharper dispersion between companies turning AI and major launches into durable cash flow versus those facing high expectations, tight funding, or margin pressure.

Important Disclaimer

​

This newsletter is for informational purposes only and does not constitute investment advice. Content is generated by AI and may contain inaccuracies; always verify data independently before trading. Investing involves significant risk of loss. AlchemyJ is not a registered financial advisor. By reading this, you agree to our terms.

​

View our Full Legal & AI Disclaimer here

TechTrends banner en.png
AlchemyJ Agentic AI Investment Signals

An AI-Powered Financial Intelligence Platform

Disclaimer:

All investment insights, sentiment analyses, and research observations provided by Alchemy
J (including but not limited to TechTrends) are generated automatically by Artificial Intelligence (AI) and automated modeling engines. While we strive for accuracy, AI-generated content is inherently subject to algorithmic limitations and may contain errors, omissions, or inaccuracies.

The content provided on this platform is for informational and educational purposes only and does not constitute financial, investment, legal, or tax advice, nor does it represent an endorsement, offer, or solicitation to buy or sell any security.

Users are strictly required to independently verify all data, metrics, and analyses before making any investment decisions. Alchemy
J, its affiliates, and its operators assume no liability or responsibility for any financial loss, damages, or adverse outcomes resulting directly or indirectly from the use of or reliance upon this platform's content. Past performance is not indicative of future results. Invest at your own risk.

©2026 AlchemyJ Pte. Ltd. All Rights Reserved.

bottom of page