Micron Secures $100 Billion AI Demand.

Mazen Al Sabbagh

— 7 min read
Share

TL;DR

  • Micron's Q3 performance: Surpassed all estimates with $41.46B revenue, $25.11 EPS, and 84.9% gross margins, projecting $50B in Q4 revenue at 86% margins. Shares surged, triggering a broader semiconductor market recovery.
  • OpenAI's custom silicon: In collaboration with Broadcom, OpenAI introduced "Jalapeno," its first custom inference chip. Developed in nine months, deployment is slated for year-end at gigawatt-scale data centers.
  • Alibaba's alleged IP infringement: Anthropic accused Alibaba's Qwen lab of conducting a large-scale adversarial distillation attack using 25,000 fraudulent accounts against Claude, disclosed in a letter to US officials. Alibaba shares declined significantly.
  • Google's talent drain: A critical talent exodus continued at Google, with five high-profile AI researchers, including Gemini 2.5 contributors, departing for OpenAI and Anthropic within a week.
  • Fable 5 restoration signal: Leaked code from Claude Code v2.1.190 indicates Anthropic is integrating weekly Fable 5 usage into subscription plans, suggesting an imminent return for the model.

Lead Story: Micron Secures $100 Billion AI Demand.

Forty-eight hours after a single Korean-language article regarding HBM4 production forecasts triggered a global AI selloff, Micron Technology delivered a definitive rebuttal. The company's fiscal Q3 results, reported after Wednesday's close, exceeded every key estimate: revenue reached $41.46 billion (vs. $35.25B consensus), adjusted EPS hit $25.11 (vs. $20.20), and gross margins stood at 84.9%—a significant increase from 39% a year prior. Revenue surged 74% sequentially and 346% year-over-year, with the core data center business expanding sevenfold to $11.5 billion.

The figures that fundamentally reshaped the structural debate, however, were in the guidance. Micron projected Q4 revenue of $50 billion with 86% gross margins and $31 EPS—all substantially above consensus. More critically, the company unveiled 16 Strategic Customer Agreements structured as take-or-pay contracts, featuring binding multi-year commitments through 2030. These agreements represent approximately $100 billion in minimum contracted revenue, supported by $22 billion in upfront deposits. The contracts encompass roughly 20% of Micron's DRAM volume and one-third of its NAND output. This is not merely a revenue beat; it signifies locked-in demand with established floor prices, directly undermining the "bubble" narrative that precipitated Tuesday's market panic.

Markets responded decisively. Micron shares surged roughly 18% after hours and traded as high as $1,255 during Thursday's session. In Seoul, the KOSPI closed up 3.26% amid a broad semiconductor recovery, with Samsung Electronics climbing 9.1% and SK Hynix—which dethroned Samsung as the KOSPI's most valuable stock after a 25-year reign—rising 2.7% following its $29.4 billion Nasdaq ADR filing. Nasdaq 100 futures had jumped 1.8% overnight. The immediate concern shifts from a question of market support to whether this constitutes a durable rebound or, as Saxo Bank posited, "just a very expensive sigh of relief."

In Other News

OpenAI Unveils Jalapeno, Its First Custom Chip.

OpenAI and Broadcom on Wednesday revealed "Jalapeno", OpenAI's inaugural "Intelligence Processor"—a custom inference accelerator designed from scratch for large language model workloads. The chip was developed in an exceptionally rapid nine months and is specifically optimized for inference rather than training. Initial testing indicates performance-per-watt "substantially better" than current state-of-the-art, likely referencing Nvidia's GPUs. Broadcom CEO Hock Tan characterized compute demand from their six customers as "simply insatiable." Deployment is scheduled by end of 2026 at gigawatt-scale data centers with Microsoft and other partners, forming part of a multi-generational compute platform. This strategic move aims to reduce OpenAI's reliance on Nvidia and signifies that the economics of inference—running models at scale for applications like Codex—are becoming the paramount cost challenge in AI, compelling labs to develop proprietary silicon solutions.

Anthropic Accuses Alibaba of Industrial-Scale AI Theft.

In a letter to the Senate Banking Committee and White House officials dated June 10, Anthropic formally accused operators linked to Alibaba's Qwen AI lab of executing the largest known distillation attack on its proprietary models. The campaign, conducted between April 22 and June 5, purportedly utilized approximately 25,000 fraudulent accounts to generate 28.8 million adversarial exchanges with Claude, specifically targeting software engineering and agentic reasoning capabilities. Anthropic cautioned that models developed through such adversarial distillation processes inherently lack robust safety guardrails. Alibaba shares fell as much as 4.8% in Hong Kong on Thursday, hitting a 16-month low and contributing to a decline in the Hang Seng Index. The timing of Anthropic's disclosure is strategic: it strengthens their national-security argument for Fable 5's restoration—implying that if Chinese actors are already extracting model capabilities through accessible channels, keeping the model offline from US users may be counterproductive.

Google's AI Talent Exodus Hits Five in a Week.

Bloomberg reported Wednesday that Jonas Adler and Alexander Pritzel, both significant contributors to Google's Gemini models, intend to depart for Anthropic. Adler was instrumental in AI coding capabilities, and Pritzel in model training systems. By midday Thursday, a fifth departure emerged: Arthur Conmy, a senior research engineer on Gemini 2.5, announced his move to Anthropic to focus on train-time alignment for upcoming models. These three exits follow Noam Shazeer (to OpenAI, June 18) and Nobel laureate John Jumper (to Anthropic, June 19)—totaling five high-profile departures within a single week. Google had paid $2.7 billion to re-secure Shazeer from Character.AI in 2024; he remained for less than 22 months. Concurrently, Google launched computer use capabilities for Gemini 3.5 Flash on Wednesday, achieving a 78.4 score on OSWorld-Verified tasks—matching Sonnet 4.6—but separately confirmed that Gemini 3.5 Pro has been delayed beyond June, missing Pichai's I/O commitment.

X / Social Pulse

The market whiplash from Tuesday's KOSPI circuit breaker to Wednesday's Micron vindication dominated financial AI discourse. Semiconductor analysts who spent Tuesday defending the AI thesis pivoted to "told you so" threads within hours of the earnings release. The Fable 5 leaked code—verified strings in Claude Code indicating weekly subscription usage—triggered renewed speculation on X regarding imminent restoration, with @blader (Siqi Chen) noting Polymarket odds had jumped from 15% to 60% after Tom Brown replaced Amodei in Commerce negotiations. Arthur Conmy's announcement of his departure from DeepMind for Anthropic added a fifth name to Google's weekly exodus, prompting threads questioning whether DeepMind is evolving into a "feeder program." Reid Hoffman's Fortune interview, where he characterized xAI as a "complete train wreck" and Cursor as "fading over the horizon," continued to circulate. Goldman Sachs data embedded in the same interview—AI erasing approximately 11,000-16,000 net US jobs per month, with Gen Z disproportionately affected—received less attention but arguably carries greater long-term significance.

One to Watch

SK Hynix filed for a $29.4 billion Nasdaq ADR listing, more than doubling its original March target of $14 billion—and on Thursday, it dethroned Samsung Electronics as the KOSPI's most valuable stock, concluding a 25-year reign. The ADRs are anticipated to commence trading July 10, led by Bank of America, Citigroup, Goldman Sachs, and JPMorgan Chase. Proceeds are designated to fund new fabrication capacity at Yongin and advanced packaging at Cheongju, alongside critical EUV equipment acquisitions—the company ordered $8 billion in ASML scanners in March alone. SK Hynix currently trades at 7.5x forward earnings in Seoul, compared to Micron's 9.5x and TSMC's 21x; the US listing is explicitly designed to narrow this valuation disparity. At its proposed size, this would constitute one of the three largest first-time share sales in history, providing global investors with direct Nasdaq access to the company controlling 57% of the HBM market essential for powering Nvidia's AI processors.

Quick Hits

  • MGX establishes $50B AI fund: Abu Dhabi's MGX secured $50 billion from investors for one of the largest dedicated AI investment vehicles, targeting $100 billion AUM and $10 billion annual deployment. (Bloomberg)
  • Cerebras' post-IPO volatility: First earnings post-IPO showed 92% YoY revenue growth to $193.4M, yet shares dropped nearly 20% following margin guidance described as "misunderstood." (CNBC)
  • Qualcomm acquires Modular: Qualcomm confirmed its $3.9 billion acquisition of Modular, integrating Mojo and MAX into its $14 billion anti-Nvidia strategy alongside the pending Tenstorrent deal. (CNBC)
  • Meta's "Arena" prediction market: Meta is developing "Arena," an AI-powered prediction market app using Llama to generate and resolve markets in real-time, impacting DraftKings and Robinhood shares. (NPR)
  • GPT-4.5 retirement imminent: GPT-4.5 will be retired from ChatGPT on Friday, June 27, concluding its 30-day sunset period. Conversations will seamlessly transition to GPT-5.5. (OpenAI)

Tuesday's market question was whether the AI infrastructure thesis constituted a speculative bubble. Wednesday's answer, delivered by Micron, manifested as $100 billion in binding contracts with established floor prices—not projections, but signed commitments. SK Hynix's $29.4 billion Nasdaq filing and its ascendancy over Samsung as Korea's most valuable stock amplified this signal: the largest memory chipmakers are not hedging against a slowdown; they are raising historic capital to expand capacity. OpenAI designing its own silicon, Anthropic leveraging distillation data as geopolitical ammunition, and leaked code indicating Fable 5's return is being engineered into subscription infrastructure all point to the same underlying reality: the AI industry has advanced beyond the phase where model benchmarks unilaterally determine market leadership. The core contest now centers on infrastructure, supply chain dominance, and political leverage. Five Google researchers departing for rivals in a single week is the talent market's unequivocal validation of this shift. The selloff is not definitively over; Micron's earnings purchased a reprieve, not a resolution. Yet, for one day, the data resonated with more authority than the doubt.

Sources

Lock in. M. mazen@thorterminal.com

Read more