January 18th

TSMC reported record Q4 net income of $16 billion, a 35% year-over-year increase, with CEO C.C. Wei stating AI demand is "endless" and the chipmaker planning to spend between $52 billion and $56 billion on capital expenditures in 2026.

Gartner forecasts global AI spending will hit $2.53 trillion in 2026 and surge to $3.33 trillion by 2027, suggesting the market is ignoring "bubble" fears as significant capital flows into data center infrastructure despite concerns over immediate return on investment.

Mercor reached a $10 billion valuation by hiring over 30,000 experts—from dermatologists to fiction writers—to generate high-quality training data for AI labs, industrializing the transfer of human expertise to machines and creating a gig economy focused on training replacements for high-skill labor.

Parloa raised $350 million at a $3 billion valuation to expand its AI customer service platform, tripling its value in just eight months and reflecting continued venture capital confidence in AI agents that can automate complex enterprise interactions.

Anthropic launched "Cowork," a desktop agent capable of reading, editing, and managing local files and folders for general computing tasks, reportedly built in under two weeks using Anthropic's own coding tools and moving AI utility beyond chat windows into direct operating system interaction.

Salesforce overhauled Slackbot into an AI "super agent" powered initially by Anthropic's Claude, capable of synthesizing data across Salesforce records, Google Drive, and Slack conversations, with internal testing showing rapid adoption among 80,000 employees and validating enterprise appetite for agents that execute complex workflows.

Apple entered a multi-year partnership with Google to power the next generation of Siri and Apple Intelligence features using Gemini models and cloud infrastructure, acknowledging the immense capital and compute required to maintain frontier models and positioning Google as the infrastructure backbone for billions of iOS devices.

The Pentagon announced plans to integrate xAI's Grok and Google's Gemini into military networks this month to analyze data across classified and unclassified systems, highlighting a divergence between military utility priorities and ongoing civilian investigations into xAI.

Anthropic released data showing that its Claude model is predominantly used for high-skill tasks, warning that removing these tasks from human workflows could lead to the "deskilling" of professions like technical writing and travel planning, while roles like property management may see "upskilling" as routine bookkeeping is automated.

The IMF reported that job postings requiring new skills pay a premium of up to 15% in the UK and 8.5% in the US, but employment in AI-vulnerable occupations is 3.6% lower after five years in high-AI-demand regions, indicating AI is already reshaping labor demand and potentially reducing entry-level hiring.

Goldman Sachs research warned that AI could potentially automate tasks accounting for 25% of all work hours in the US, with the rapid capability growth of agents like Anthropic's Cowork suggesting a looming inflection point for white-collar automation.

California Department of Justice launched a formal investigation into xAI after its Grok chatbot was used to generate nonconsensual, sexualized deepfakes of women and children, joining a growing international backlash that includes a formal investigation by the UK's Ofcom and service blocks implemented by Indonesia and Malaysia.

The US Senate passed the DEFIANCE Act, which would establish a federal civil cause of action allowing victims of nonconsensual sexually explicit deepfakes to sue creators and distributors, representing a concrete step toward closing legal loopholes that currently shield platforms from liability.

ETSI released the EN 304 223 standard, the first globally applicable European Standard for AI cybersecurity, establishing baseline requirements for securing AI models against data poisoning and model obfuscation while defining clear roles for developers, system operators, and data custodians.

Google removed AI Overviews for certain medical searches following a Guardian investigation that revealed the tool was providing "dangerous and alarming" health information, underscoring the persistent reliability gap in using LLMs for high-stakes domains like healthcare.

Microsoft pledged to cover the full costs of procuring renewable energy for its data centers to prevent residential electricity rates from rising in host communities, addressing growing public backlash against the massive energy and water consumption of AI facilities that threatens to stall infrastructure expansion.

PC manufacturing costs for memory and storage rose 40% to 70% in 2025 due to AI data center demand, causing a severe RAM shortage that is forcing OEMs to lower specs or raise prices and effectively cooling the marketing hype around consumer "AI PCs" that require high memory for local processing.

OpenAI launched a Request for Proposals (RFP) to stimulate US-based manufacturing for key AI supply chain components, including data center inputs and robotics, aiming to reduce reliance on foreign supply chains and strengthen domestic capacity as infrastructure needs scale.

OpenAI, Google, and Anthropic all launched specialized healthcare tools within days of each other, though none have FDA clearance for diagnosis, highlighting a race to capture the medical market despite unresolved liability questions and regulatory approval processes.

Google debuted "Personal Intelligence" for Gemini, a beta feature that reasons across a user's Gmail, Photos, and Drive to provide proactive, personalized answers, leveraging its ecosystem dominance to create a utility moat against standalone chatbots by connecting disconnected data silos.

OpenAI launched "ChatGPT Go," a lower-cost subscription tier at $8/month, expanding access to its models globally in a pricing strategy aimed at capturing price-sensitive users in international markets and countering the rise of efficient, lower-cost models from competitors.