Episode Details

Meta's Muse launch turned a consumer AI agent into a CPU demand story, a new round of frontier model releases sharpened the fight over token economics, and AMD crossed the trillion-dollar mark on the same compute thesis. Patrick Moorhead and Daniel Newman break down Meta Connect, Qualcomm's agentic-device push, the AI policy debate, and whether Meta has finally cracked mainstream AI wearables.

The handpicked topics for this week are:

1. Meta Connect Turns Muse Into a Cross-Device Agent Platform: Meta used its developer conference to extend Muse, its consumer AI agent, across new Ray-Ban audio glasses, a camera-equipped version, and a palm-sized Muse Charm device. Patrick Moorhead tested Muse himself, booking and updating appointments and pricing airline tickets, and called its computer-use capability the real unlock. Daniel Newman pointed to a download pace faster than ChatGPT's and roughly 700,000 daily users, and argued Meta's bigger opportunity is taking a small cut of the transactions agents complete, since only 2% to 3% of consumers pay for AI today. Newman added that Futurum had raised its 2030 forecast for AI CPU revenue from $110 billion to $250 billion days before the launch, and expects to revise it again. (The Decode)

2. Qualcomm Builds Out the Agentic-Device Stack at Snapdragon Summit: In Maui, Qualcomm moved from one flagship smartphone chip to two and introduced an audio chip that lets earbuds and glasses connect over Wi-Fi without a phone, a move Daniel Newman read as a direct shot at Apple. Newman also flagged a new seven-year patent license with Apple and the acquisition of a robotics software company built around the open-source MoveIt platform. Patrick Moorhead focused on how Qualcomm's core technologies, including HBC stacking and modular designs, scale from wearables to hyperscale data centers, arguing that volume across billions of chips could give Qualcomm a cost advantage despite a projected 1% to 2% data center share in 2029. (The Decode)

3. Opus 5.5, GPT-6 Sol and Luna, and Grok 4.7 Intensify the Model Race: Anthropic, OpenAI, and xAI all shipped new models within days of each other. Patrick Moorhead cited Signal65 research he says shows Anthropic's newest model using roughly three times the output tokens of its top rivals to reach correct answers, which he tied to Anthropic's compute constraints, and noted OpenAI cut Sol pricing by up to 50%. Daniel Newman shared that one Fortune 100 company spending $200 million a year with Anthropic plans to cut that by about 75% by shifting workloads to open-source and on-device models. Newman argued the race is far from settled, with OpenAI, Meta, Google, and xAI all showing momentum, and that the labs that locked in compute early are best positioned to stay on the frontier. (The Decode)

4. AI Alarmism Meets Policy as Jensen Huang Rejects Extinction Risk: The push to rename AI "superintelligence" and a proposal from Sen. Bernie Sanders to ban superintelligence development kept the policy debate heated, while Jensen Huang said there is zero chance of AI causing human extinction. Daniel Newman warned that viral safety headlines rarely get corrected once they spread. Patrick Moorhead argued the tech industry lost control of the message by emphasizing job loss and danger without making the case for AI's benefits or explaining what losing the race to China would actually mean. (The Decode)

5. U.S.-China Talks Produce Pageantry but No Chip Agreements: President Xi's U.S. visit brought parades and a high-profile dinner with technology leaders, but Patrick Moorhead noted no agreements on chips or chip exports. He pointed to a two-month extension announced by Treasury Secretary Scott Bessent and farm purchase commitments, with Taiwan at the top of China's ag