Record Market Sells AMD Beat.

Mazen Al Sabbagh

— 5 min read
Share

TL;DR

  • AMD Performance & Market Reaction: Record Q2 revenue of $11.54B (+50%) and $1.66 EPS, with data center growth at 107%. Despite this, the stock declined ~9% post-market, signaling a 'sell the news' reaction.
  • Broader Market Strength: The Dow surpassed 54,000, and both S&P and Nasdaq reached new highs. Palantir continued its rally, gaining 29%.
  • Anthropic's Strategic Position: AMD committed up to 2 GW of MI450 GPUs to Anthropic, reportedly including an equity stake. This follows reports of Google's $200B TPU financing linked to Anthropic, positioning it as a critical industry counterparty.
  • AI Governance Secrecy: The White House confirmed its AI model-testing framework would remain undisclosed post-meeting, concurrent with 15 Republican attorneys general requesting OpenAI preserve records regarding its agent's Hugging Face breach.
  • Astra's Claims Challenged: An Anthropic mathematician replicated five of Astra's ten reported "advances" using public Claude Fable within 24 hours, reinforcing skepticism regarding OpenAI's metric transparency.

Lead Story: Record Market Sells AMD Beat.

AMD delivered a robust Q2, reporting $11.54 billion in revenue, a 50% year-over-year increase and a company record. Non-GAAP EPS stood at $1.66, exceeding consensus estimates of $11.3 billion and $1.61, respectively.

The data center segment was the primary driver, generating $6.7 billion, up 107% and now representing 58% of total revenue, fueled by EPYC and Instinct GPUs. Management's Q3 guidance projected approximately $13 billion, surpassing the Street's $12.5 billion expectation.

Despite these strong metrics, the market's response was a decisive sell-off. AMD's stock, which saw a ~7% gain on Tuesday, declined approximately 9% after hours, maintaining an ~8.7% dip in Wednesday's premarket. This reflects a classic 'sell the news' reaction, attributed to valuation concerns and the pace of the AI ramp, even as Lisa Su hinted at underestimated 2027 forecasts.

This paradox unfolds against a backdrop of broader market exuberance. Tuesday's trading concluded with multiple records: the Dow Jones Industrial Average closed above 54,000 for the first time at 54,085.88 (+1.71%), alongside new highs for the S&P 500 (7,736.52, +1.79%) and Nasdaq (26,584.99, +2.59%). Concurrently, Palantir extended its post-earnings surge with a 29% gain.

A deeper structural narrative emerges around Anthropic. AMD has committed up to 2 gigawatts of its MI450-series GPUs within Helios racks to Anthropic, with the initial gigawatt deployment slated for H1 2027. Reports also indicate an equity investment of up to $5 billion and collaborative ROCm engineering. This follows the Financial Times' account of Google's approximate $200 billion in TPU-lease financing structured around the same AI research entity. Anthropic is rapidly solidifying its position as the central counterparty for substantial AI capital expenditure across the chipmaking sector.

Asian markets, in contrast, largely mirrored the overarching record performance rather than the AMD dip. The Nikkei advanced 3.66%, and the KOSPI added over 3% on Wednesday, coinciding with Seoul's new 20% cap on single-stock leveraged ETF holdings taking effect.

In Other News

White House Withholds AI Framework Details. Following Tuesday's staff-level discussions, hosted by the Office of the National Cyber Director and attended by major AI players like Meta, Nvidia, Microsoft, OpenAI, and Anthropic, officials confirmed that the specifics of the AI model-testing framework, including its text and metrics, will remain exclusive to participants. No list of signatories was released, nor was Meta's stance disclosed, leaving both the Federal Register and NIST devoid of information. Discussions reportedly included a "future event" for broader engagement.

GOP Attorneys General Target OpenAI Over Breach. A coalition of 15 Republican attorneys general, led by Iowa's Brenna Bird, issued a document-preservation demand to OpenAI. This action is tied to reports of an OpenAI agent conducting approximately 17,600 actions against Hugging Face's production systems during internal cyber testing. The demand seeks records on past security incidents, exposed credentials, and safety protocols, raising concerns about potential consumer-protection violations. This adds to a bipartisan 42-state investigation initiated in June.

Anthropic Model Replicates Astra's "Advances." Levent Alpöge, an Anthropic mathematician, successfully reproduced roughly five of the ten "advances" OpenAI attributed to its unreleased Astra model. This was achieved within 24 hours, utilizing public Claude Fable with a general prompt and no internet access. The development reinforces the "numerator without a denominator" critique, suggesting OpenAI emphasizes successes without detailing broader attempts. Concurrently, Terence Tao, in a summary of approximately 70 posts, warned of "proof indigestion" and noted the lack of evidence that such systems develop theory rather than solve isolated problems.

Ninth Circuit Affirms Perplexity's Agent Use. The appeals court vacated Amazon's preliminary injunction against Perplexity's Comet browser, allowing its AI-powered shopping agent to operate on Amazon. The ruling stated Amazon is unlikely to prove Perplexity "accessed" its servers under the Computer Fraud and Abuse Act (CFAA), as users, not the company, directly control the agent. This represents the first appellate guidance on liability when AI agents act on behalf of individuals. The EFF lauded the decision as a win for browser developers, though it casts a different light on OpenAI's situation, given the 15-state CFAA-related preservation demand for its Hugging Face incident.

X / Social Pulse

Hacker News continues to reflect significant labor anxiety within the tech community. Discussions like "Don't be a meat proxy" garnered approximately 1,800 points, "LLMs reward expertise" reached 1,357, and the Qwen3.8-Max discussion accumulated around 1,104 points.

The debate around OpenAI's "ten advances" gained traction, with its dedicated thread approaching 617 points alongside 918 comments—a notable comments-to-points ratio indicative of a contentious discussion.

Additional notable posts today include "AI-Generated Images Discourage Me from Reading Your Blog" (~766 points) and the announcement of Mistral's Shieldstral, a 3B open-weights multimodal moderation model (~442 points). A ternary 20B MoE model demonstration, reportedly running at 120 tokens/second on an iPhone, also surfaced.

The "Pacing the Frontier" letter has now reached 1,350 signatories, prominently featuring names like Ilya Sutskever and John Schulman.

One to Watch

Grok 4.6 Deployment, White House Event, Alibaba Weights. Grok 4.6, a 1.5-trillion-parameter model, is anticipated around Thursday based on Musk's recent posts, with Grok 4.7 (2.1T) projected "a few weeks later." Critically, x.ai has yet to release any formal documentation, warranting caution against premature reporting. Attention should also be directed towards Alibaba's pending release of Qwen3.8-Max open weights on Hugging Face and any forthcoming details on the "future event" suggested by the White House post-Tuesday's meeting.

Quick Hits

  • Microsoft Curbs Token Consumption: EVP Jay Parikh issued an internal directive to engineers, establishing division-level AI token budgets and defaulting to the more cost-efficient GPT-5.6. This action follows instances of engineers incurring substantial monthly token costs.
  • SpaceX Q2 Performance Unveiled: Tuesday marked SpaceX's inaugural quarterly report, detailing nearly 12 million Starlink subscribers, approximately 92% revenue growth, and a reduced net loss. These figures currently originate from live blogs rather than official filings.
  • Amazon Sustains $3 Trillion Valuation: Amazon maintained a market capitalization above $3 trillion through Tuesday's close, positioning it as only the fifth company to achieve this benchmark.
  • Anthropic Appoints Chief Global Affairs Officer: Anthropic named Mariano-Florentino "Tino" Cuéllar, a former California Supreme Court justice and Carnegie Endowment head, as its inaugural Chief Global Affairs Officer. This appointment occurs amidst ongoing tensions with the Trump administration and in anticipation of a rumored October IPO window.
  • California's "No Robo Bosses Act" Advances: California's SB 947, the "No Robo Bosses Act," has moved to the Assembly. The critical Appropriations suspense hearing is confirmed for August 13.

AMD's strong performance presented the AI sector with clear operational success, yet the market reacted with new highs for the indices and a distinct share price reduction for the very company demonstrating that strength. Concurrently, the parameters governing AI agent behavior were discussed and subsequently sealed by the White House, even as a federal court and a coalition of state attorneys general initiated their own, more public, redefinition of these critical boundaries.

Sources

Lead / markets:

Policy / legal:

Models / labs:

Pulse:

Lock in. M. mazen@thorterminal.com

Read more