Market Overview

Anthropic signed a $9.1 billion computing deal with Riot Platforms, a Bitcoin miner, to supply 191 megawatts of capacity from Riot's Rockdale, Texas facility over 20 years through June 2048. The contract could reach $16.1 billion if both five-year extension options are exercised. Riot will receive long-term, predictable revenue from its existing power and cooling infrastructure instead of mining Bitcoin, which generates volatile returns tied to crypto prices. Anthropic is racing to lock in compute capacity as AI model training demands explode; it has also committed to $10 billion with Volta Infra and roughly $45 billion with xAI. Riot's stock surged 26% overnight and shifted to "extremely bullish" retail sentiment, though the company swung to a $237.2 million net loss in Q2 despite 14% revenue growth to $174.2 million. The deal reflects a broader pivot by former miners—Core Scientific, TeraWulf, IREN, and others—to monetize scarce power assets for AI and high-performance computing, where margins and contract certainty exceed Bitcoin mining.

Trump Media reported a $238 million net loss for the second quarter, more than ten times the $20 million loss in the same period last year, driven almost entirely by a $361 million decline in the value of its cryptocurrency holdings. The company held $557 million in bitcoin as of June 30, down from $836 million at year-end, and had pledged roughly a significant amount of bitcoin as collateral against convertible debt and options strategies. Revenue climbed 89% to $1.7 million, mostly from Truth Social advertising, but operating expenses jumped 275% year-over-year to $165 million, partly because the company marks its digital assets to market each quarter. The losses prompted new CEO Kevin McGurn to unwind two major crypto partnerships with Crypto.com and Yorkville Acquisition, scrapping plans for a publicly traded CRO treasury company and scaling back prediction-market integration into Truth Social. Instead, McGurn is betting on two revenue plays: a new API service called Truth API that sells faster access to Trump's posts to high-frequency trading firms at $60,000 to $100,000 monthly (now with more than 10 customer agreements), and a pending merger with TAE, a fusion energy company McGurn called "the single most important driver of long-term value." The company has never turned a profit and remains a crypto holdings vehicle wrapped around a media business, according to analyst Markus Thielen, though McGurn is attempting to reposition it as a data and distribution platform for financial services.

Meta released Muse Glimmer, a 30-billion-parameter open-weight AI model designed to run on consumer laptops, and said it would open-source the weights of Muse Spark 1.2, its most powerful model to date. CEO Mark Zuckerberg used the launch to argue that U.S. policy must reduce restrictions on training data and model distillation if American firms are to compete with Chinese open-weight systems from DeepSeek, Moonshot, and Alibaba, which already rival top U.S. models in performance. The move reflects a strategic reset: Meta's frontier models have lagged behind OpenAI and Anthropic in enterprise adoption and revenue, and the company is repositioning itself as a cheaper, customizable alternative rather than chasing the same closed-model strategy as its rivals. Glimmer runs locally on a single GPU, avoiding cloud compute costs and addressing real demand from businesses spooked by ballooning AI bills and recent security breaches tied to closed-source systems. Zuckerberg also announced a governance change giving independent directors power to approve safety criteria before model releases, a concession to investor and regulatory concern. The stock rose in premarket trading, though Meta shares remain down for the year as Wall Street questions whether the company's capex forecast will generate returns. Advertising still accounts for 97.7% of Meta's revenue, and the company is betting AI will sharpen ad targeting further while also creating new revenue streams from selling spare data center capacity. The open-weight bet is a retreat from Meta's earlier ambition to compete head-to-head with frontier labs on closed models, but it positions the company to capture demand from developers and enterprises that prefer domestic alternatives to Chinese systems and want to avoid vendor lock-in.