This week in AI, the industry is grappling with a fascinating tension: unprecedented infrastructure investment and hardware innovation collide with growing skepticism about AI's true value and the psychological toll it's taking on the tech ecosystem. SoftBank's staggering €75 billion pledge for French data centers underscores the relentless build-out, while startups like Groq and Xcena raise hundreds of millions to challenge Nvidia's dominance. Yet, a parallel narrative has emerged around "AI psychosis" — a term capturing the industry's collective delusion about AI's capabilities. Developer backlash against GitHub Copilot's new billing model and warnings from Cognition's CEO about over-reliance on coding agents suggest the honeymoon phase may be ending. Meanwhile, Meta reportedly doubles down on wearables, and Erin Brockovich enters the fray, questioning data center secrecy. It's a week of big bets, big questions, and a growing demand for accountability.
In a jaw-dropping display of capital commitment, SoftBank announced plans to invest up to €75 billion to construct AI data centers across France. This massive bet signals that the infrastructure arms race is far from over, with SoftBank positioning itself as a key player in Europe's AI sovereignty ambitions. The investment, one of the largest single-country pledges in tech history, will likely reshape the European cloud and AI compute landscape for years to come.
A provocative term is taking hold in Silicon Valley: "AI psychosis." This week, the debate exploded, with Box CEO Aaron Levie arguing that most CEOs are suffering from it — a dangerous overconfidence in AI's current capabilities that leads to bad strategic decisions. The concept describes a collective delusion where executives, investors, and technologists project human-like intelligence onto systems that are fundamentally pattern matchers, leading to overinvestment and under-delivery. It's the most important psychological diagnosis the industry has faced since the "fake it till you make it" era of Web 2.0.
GitHub has ignited a firestorm by switching Copilot from a flat monthly fee to a token-based billing model, and developers are not happy. The change, which charges per token generated, has been widely mocked on social media as a money grab that introduces unpredictable costs for what many view as a productivity tool, not a utility. This backlash highlights a growing friction point: as AI tools become essential to workflows, their pricing models are becoming increasingly contentious.
Hot on the heels of Nvidia's massive "not-acqui-hire" deal, AI chip startup Groq is reportedly in the process of raising a $650 million funding round. The news underscores the voracious appetite for alternatives to Nvidia's dominant GPU architecture, especially for inference workloads. Groq's unique LPU (Language Processing Unit) architecture is gaining traction as enterprises seek more efficient and specialized hardware for running large language models in production.
Environmental activist Erin Brockovich has turned her attention to the AI industry, specifically targeting the secrecy surrounding data center construction and operations. Brockovich is raising questions about water usage, energy consumption, and the lack of community transparency from major tech companies building massive facilities. Her involvement signals a new front in the AI accountability movement, one that moves beyond algorithmic bias to address the physical and environmental footprint of the AI boom.
A growing cohort of developers is refusing to take jobs that don't provide AI coding assistants, viewing them as essential modern tools. But industry veterans are warning that this dependency is creating a dangerous skills atrophy, where junior developers never learn fundamental debugging, architecture, or systems thinking. Cognition CEO Scott Wu, whose company builds the AI coding agent Devin, even cautioned that these tools should augment, not replace, human developers — a rare note of restraint from a founder in the space.
Meta is doubling down on wearable AI, reportedly developing an AI-powered pendant that would serve as a always-on, voice-activated assistant. The device, which would compete with the likes of Humane's AI Pin and Rabbit's R1, represents Meta's continued push into hardware beyond the Quest headset and Ray-Ban smart glasses. The key question remains whether consumers are ready for a dedicated AI wearable, or if this is another solution in search of a problem.
Xcena, a chip startup focused on memory architecture, has raised $135 million at a $570 million valuation, betting that the real bottleneck in AI isn't processing power but memory bandwidth. As AI models grow exponentially larger, the ability to move data between memory and compute units has become a critical constraint. Xcena's thesis — that solving memory latency and bandwidth will unlock the next generation of AI performance — is gaining serious investor attention and could challenge the prevailing "more GPUs" orthodoxy.
In a hands-on review, TechCrunch's reporter put Google's always-on AI assistant, Gemini Spark, through its paces and came away impressed. Unlike previous digital assistants that felt gimmicky, Gemini Spark demonstrated real utility for task management, scheduling, and information retrieval across a user's digital life. The review suggests that Google may have finally cracked the code on a truly useful AI assistant, though questions about privacy and the "always listening" nature of the device remain.
The browser wars are back, and AI is the new battleground. A new crop of browsers — including Arc, SigmaOS, and Brave — are differentiating themselves through built-in AI features like automatic tab grouping, smart summarization, and AI-powered ad blocking. These challengers are forcing Chrome and Safari to play catch-up, creating a genuinely exciting moment for a product category that had grown stale. The piece highlights how AI is becoming a core feature, not just an add-on, for the next generation of web browsers.
TechCrunch's video segment this week explored what happens when companies become "too AI-pilled" — a slang term for being overly intoxicated by AI hype. The piece argues that companies that rush to replace human judgment with AI output, slash their workforces prematurely, or build products around AI capabilities rather than customer needs are setting themselves up for spectacular failure. It's a necessary reality check for an industry that sometimes forgets that technology serves business, not the other way around.
For those drowning in AI jargon, TechCrunch published a much-needed glossary explaining terms like "hallucination," "fine-tuning," "RAG," and "AGI" in plain English. As AI literacy becomes a professional necessity, this guide serves as a valuable reference for anyone who's nodded along in a meeting without truly understanding what was being said. It's a sign that the industry is maturing — finally acknowledging that clear communication matters.
In a thought-provoking piece from MIT Technology Review, the Pope's encyclical "Magnifica Humanitas" is examined as a philosophical framework for navigating the AI era. The document argues that human dignity, creativity, and moral responsibility must remain central as we integrate AI into our lives. It's a refreshingly humanistic take that offers a counterpoint to the techno-optimism and techno-pessimism that dominate most AI discourse.