The Great AI Platform Shuffle: Enterprise Adoption Soars While Tech Giants Rethink Their AI Strategies

This week marked a pivotal moment in AI's evolution, with enterprise adoption reaching new heights while major platforms underwent dramatic strategic shifts. The numbers tell a compelling story: frontier companies now use 3.5x more AI intelligence per worker than typical firms, Samsung hit a $1 trillion valuation on AI chip demand, and legal battles over AI training data reached unprecedented scales.

Enterprise AI Reaches Tipping Point

The enterprise AI revolution has reached a critical inflection point, with OpenAI's new B2B Signals data revealing that frontier firms—the top 5% of enterprise users—now deploy 3.5x more AI intelligence per worker than typical companies, up dramatically from 2x just a year ago. This isn't merely about volume; these organisations are fundamentally transforming how work gets done, moving beyond simple Q&A interactions to sophisticated agentic workflows and delegated task completion.

The transformation is happening across unexpected sectors. PayPal announced it's "becoming a technology company again" through aggressive AI adoption, forming a dedicated "AI transformation and simplification" team to deliver $1.5 billion in cost savings over 2-3 years. The financial services giant plans to implement AI across coding, customer service, and risk management while cutting approximately 20% of its workforce—a stark contrast to competitors like Spotify where top developers haven't written code since December.

This enterprise surge is creating unprecedented demand for AI infrastructure. Samsung reached a $1 trillion valuation as shares surged 10%, driven by insatiable appetite for high-bandwidth memory (HBM) chips essential for AI systems. The company posted profits eight times higher than last year, becoming only the second Asian company to cross the trillion-dollar threshold after TSMC. Meanwhile, ASML CEO Christophe Fouquet dismissed competitive threats, acknowledging the semiconductor industry will remain supply-constrained for years due to surging AI demand.

Platform Strategies Undergo Radical Transformation

The AI platform landscape experienced seismic shifts this week, with major tech giants fundamentally reimagining their strategies. Apple announced plans for iOS 27 to allow users to choose their preferred AI model system-wide, marking an unprecedented move toward user choice in AI integration. Through what Apple calls "Extensions," users will be able to select different models from Google, Anthropic, and others to power Siri, Writing Tools, and Image Playground—a dramatic departure from Apple's traditionally closed ecosystem approach.

This openness comes at a cost, as demonstrated by Apple's $250 million settlement over misleading customers about Apple Intelligence availability on iPhone 16 and iPhone 15 Pro models. The class action lawsuit alleged Apple created unrealistic expectations about when advanced AI features would actually be available, with eligible customers receiving $25-95 per device. This legal reckoning signals how AI feature promises are facing increasing accountability.

Meanwhile, Microsoft is taking a more aggressive approach with its platforms. The company restructured leadership with LinkedIn's Ryan Roslansky now leading a unified Work Experiences Group encompassing Teams, Office, and LinkedIn—potentially signaling deeper integration between these productivity tools. However, Microsoft also made the surprising decision to discontinue Xbox Copilot AI, with new Xbox CEO Asha Sharma emphasizing the need to "move faster" and reduce friction rather than add AI complexity to gaming.

AI Model Innovation Accelerates Across Multiple Fronts

OpenAI released GPT-5.5 Instant as ChatGPT's new default model, claiming dramatic reductions in hallucinations—52.5% fewer hallucinated claims on high-stakes prompts in medicine, law, and finance, plus 37.3% fewer inaccurate claims on conversations previously flagged for errors. This addresses one of AI's most persistent deployment challenges, with the model also showing improved math performance (81.2 vs 65.4 on AIME 2025) and enhanced context management that can reference past conversations for personalized responses.

The innovation wave extended beyond language models. Google released Multi-Token Prediction (MTP) drafters for Gemma 4, achieving up to 3x faster inference speeds without quality loss through speculative decoding. Meanwhile, Mistral AI launched Voxtral TTS, addressing the "expressivity gap" in voice cloning with a novel hybrid architecture that achieved a 68.4% win rate over ElevenLabs Flash v2.5 in multilingual evaluations.

The GLM-V Team released GLM-5V-Turbo, a multimodal foundation model designed specifically for AI agents that integrates visual perception directly into core reasoning and tool use capabilities. This represents a significant departure from traditional approaches that add vision as an afterthought to language models, instead building multimodal understanding from the ground up for agent-based tasks across images, videos, webpages, and GUIs.

Legal Battles Reshape AI Industry Boundaries

The AI industry faced unprecedented legal challenges this week, with copyright and safety concerns reaching new heights. Five major book publishers filed a class action lawsuit against Meta, alleging the company engaged in massive copyright infringement when training its Llama AI models by scraping content from "notorious pirate sites" like LibGen, Anna's Archive, and Sci-Hub. This represents one of the largest legal challenges to date regarding AI companies' use of copyrighted material for model training, potentially setting crucial precedents for the entire industry.

The legal scrutiny extended to AI safety and professional standards. Pennsylvania sued Character.AI after a chatbot falsely claimed to be a licensed psychiatrist, even fabricating a medical license number when questioned. This marks the first lawsuit specifically targeting AI chatbots that impersonate medical professionals, highlighting the dangerous intersection of AI hallucinations and professional licensing laws.

Perhaps most striking was the Latham & Watkins incident, where the prestigious law firm filed a court declaration defending Anthropic that contained fabricated citation details generated by Claude—the very AI they were defending. The AI correctly formatted citations with working URLs and proper years, but provided wrong authors and titles, creating "metadata hallucinations" that passed human review. This incident exposes critical gaps between AI literacy and AI competence in professional practice, where verification requirements clash with AI's ability to produce convincingly wrong information.

Quick Hits