AI Capex Doubts Fuel Chip Selloff.

Mazen Al Sabbagh

— 7 min read
Share

TL;DR

  • Market Correction: Global markets experienced a sharp correction as AI infrastructure investment doubts intensified, with the KOSPI activating a circuit breaker following a 10% decline driven by memory-chip concerns. The Nasdaq Composite retreated 2.2%, Micron Technology fell 13%, and the Philadelphia Semiconductor Index registered a 7.3% drop.
  • Oracle Restructuring: Oracle disclosed a reduction of 21,000 employees, or 13% of its workforce, over the past year, explicitly attributing these shifts to the integration of AI technologies. This workforce recalibration coincided with a $55.7 billion capital expenditure surge for AI data centers.
  • Frontier Model Delays: Key next-generation AI models, GPT-5.6 and Gemini 3.5 Pro, have both seen their release schedules pushed to mid-July and beyond June, respectively. This extends the current hiatus in new flagship model deployments.
  • AI Policy Funding Test: The most significant AI-funded congressional primary in history concluded with a loss for the AI safety-aligned candidate, Alex Bores, despite $13 million in PAC support. This outcome challenges the immediate efficacy of AI-centric political spending in shaping regulatory outcomes.
  • Anthropic's Enterprise Pivot: Anthropic introduced Claude Tag, an always-on AI agent for Slack, embedding Claude deeper into enterprise workflows. This strategic product expansion occurs amidst persistent issues with its Fable 5 model and a recent service outage.

Lead Story: The AI Trade's Structural Stress Test

The sustained momentum behind AI equities encountered significant resistance on Tuesday. The initial catalyst materialized in Seoul, where a Chosun Biz report indicated a downward revision in Nvidia's Rubin chip production forecasts, consequently reducing the immediate impetus for HBM4 memory adoption. This analysis led to substantial declines for SK Hynix and Samsung, pivotal components of the KOSPI, each falling over 12%. The KOSPI subsequently dropped 9.99%, triggering a 20-minute trading halt.

This regional apprehension quickly reverberated across global markets. The Nasdaq Composite declined 2.2% and the S&P 500 lost 1.4%, reflecting increasing investor skepticism regarding the long-term return on the combined $725 billion hyperscaler AI capital expenditure. Micron Technology plummeted 13% and SanDisk fell 12%, despite their prior strong performance this year. Nvidia recorded a 3.2% loss. The Philadelphia Semiconductor Index (SOX) ended down 7.3%, marking its sharpest single-day decline of 2026. Nigel Green of deVere Group aptly observed to Reuters, "The AI trade became one of the most crowded trades in global markets. When everybody owns the same stocks, the exit door becomes very small very quickly."

This market re-evaluation sets a critical stage for Micron's fiscal Q3 earnings, due after market close. Analysts anticipate approximately $20.20 EPS on $35.5 billion in revenue, representing a 967% year-over-year earnings surge. The crucial determinant will be whether Micron's guidance re-validates the AI infrastructure demand thesis or confirms the market's burgeoning anxieties.

In Other News

Oracle Cites AI in 21,000 Job Reductions. Oracle's recent 10-K filing revealed a workforce contraction of approximately 21,000 employees—a 13% reduction—over the past year. Notably, the company explicitly stated that "the adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce." This workforce recalibration incurred $1.84 billion in severance costs, a substantial increase from the prior year. Simultaneously, Oracle allocated $55.7 billion to capital expenditures—a 162% surge—to construct AI data centers for major clients. The company also intends to raise an additional $50 billion in debt and equity. Forbes highlighted that AI now accounts for an estimated 87,714 tech layoffs year-to-date, positioning it as the primary driver of job reductions in 2026.

GPT-5.6 and Gemini 3.5 Pro Ship Dates Recede. The anticipated dual-release sprint for next-generation frontier models has been postponed. Leaker @synthwavedd reported that GPT-5.6 will not be released this week, with its launch pushed to mid-July. Concurrently, Google DeepMind is reportedly dissatisfied with the current state of Gemini 3.5 Pro, delaying its release beyond June, thereby contradicting Pichai's prior commitment. A potential compensatory development is the forthcoming release of Bidi, OpenAI's new voice model, within ChatGPT this week. Anthropic's Claude Sonnet 5 is reportedly in enterprise early access as a stop-gap, given the stalled progress on Fable 5 restoration. These cumulative delays leave the frontier model landscape without a new flagship release from any major laboratory this month.

AI Funding Fails First Electoral Test. The most expensive AI-funded congressional primary in history concluded with a defeat for the AI safety advocacy. Alex Bores, a former Palantir executive who campaigned on stricter AI regulation, lost to Micah Lasher in New York's 12th District. Public First, an AI safety super PAC, committed $13 million in support of Bores, while pro-industry groups deployed substantial counter-funding, elevating total AI-linked spending in this single race beyond $40 million. Bores's concession speech underscored the industry-funded opposition. Lasher's stronger performance in older demographics suggests that while AI money can significantly influence campaigns, it has yet to galvanize the broader electorate as a decisive ballot issue.

X / Social Pulse

Developer discourse primarily centered on the GPT-5.6 delay, yet a more consequential signal emerged from the same leak thread: Claude Sonnet 5 is reportedly already in the hands of select enterprise customers via an early access program, strategically positioned as a stop-gap amid stalled Fable 5 restoration efforts. This suggests Anthropic is pursuing a two-track strategy, deploying its next model tier to paying clients while its flagship remains subject to government suspension, thereby reinforcing the July 8 identity-verification gating. Concurrently, Gizmodo published a critical analysis contending that Anthropic's ID checks "won't save Fable 5" from the export control ban. Separately, Reid Hoffman conveyed to Fortune that xAI is a "complete train wreck" due to the departure of all 11 original co-founders, and that Cursor "seems to be fading over the horizon," questioning the $60 billion valuation for a product losing market share.

One to Watch

Anthropic introduced Claude Tag on Monday—a persistent AI agent designed to reside within Slack channels, autonomously decompose tasks, and acquire organizational context over time. Both Bloomberg and Fortune characterized this as a "virtual employee" offering. The timing is strategically pertinent: with Fable 5 offline and GPT-5.6 delayed, Anthropic is shifting focus from raw model capability to expanding its product surface area. This embeds Claude more deeply into enterprise workflows, establishing a stronger competitive position before the next wave of frontier models arrives. Claude Tag will replace the existing Claude in Slack application (scheduled for sunset on August 3) and is available to Enterprise and Team customers with introductory launch credits. Anthropic reports its internal product team now utilizes Claude Tag to ship 65% of its code.

Quick Hits

  • Qualcomm is nearing a $4 billion acquisition of AI startup Modular (Bloomberg), marking its second significant AI-related deal in weeks, following the $8-10 billion Tenstorrent pursuit, underscoring a concerted push to challenge Nvidia in AI infrastructure.
  • OpenAI fully released GPT-5.5-Cyber on Sunday, achieving an 85.6% score on CyberGym, outperforming Anthropic's Mythos 5 (83.8%) on the same benchmark that precipitated export controls. Axios noted the White House's silence, raising questions regarding a potential double standard.
  • A Nature Medicine study published Monday indicated that GPT-5.2, Gemini 3.1 Pro, and Claude Opus 4.6 all surpass FDA-cleared clinical AI tools in real-world physician query performance, exposing a critical regulatory validation gap.
  • Micron and Anthropic announced a strategic partnership ahead of Micron's earnings report: Micron will serve as a primary supplier of memory and storage chips and will invest in Anthropic, solidifying AI infrastructure supply through the decade.
  • Claude experienced another outage Monday, manifesting as elevated error rates across multiple models. This marks the second notable incident in three weeks, ultimately resolved by evening.

The AI trade's structural integrity encountered a significant stress test today, not through a policy shock or a model failure, but through a re-evaluation of fundamental market incentives. A Korean-language article concerning HBM4 production forecasts served as the initial catalyst, triggering a circuit breaker in Seoul, a 7.3% decline in the Philadelphia Semiconductor Index, and a broad evaporation of over $1 trillion in market value. Oracle's explicit linkage of 21,000 layoffs to AI integration underscored the human capital shifts inherent in this technological transition. Concurrently, the delays of GPT-5.6 and Gemini 3.5 Pro left the frontier model landscape without a new flagship release on the immediate horizon. And in Manhattan, over $40 million in AI-linked political funding yielded an outcome that satisfied few: the pro-regulation candidate lost, yet the issue itself failed to resonate significantly with voters. Microsoft CEO Satya Nadella, in a Wall Street Journal interview, captured the day's disposition: "You cannot hand the world's curiosity to a handful of companies and call it progress." Micron's earnings call tonight will either restore market confidence in the AI infrastructure thesis or further validate that the prevailing valuations have outpaced the underlying reality. The AI industry has spent two years articulating a narrative of inevitable demand. Today, for the first time in months, the market ceased its unqualified affirmation.

Sources

Lock in. M. mazen@thorterminal.com

Read more