Weekly AI Roundup: The 7 Biggest Stories This Week

The last week of March 2026 delivered a packed AI news week with major model releases, regulatory action, and market-shaking spending data. Here are the seven stories that mattered most.

1. OpenAI Launches GPT-5.4 With 1M Token Context Window

OpenAI released GPT-5.4, its most capable model to date. The headline feature is a 1 million token context window in the API, which can process roughly 750,000 words in a single request. Pricing came in at $2.50 per million input tokens, a 50% cut from GPT-5. OpenAI also launched ChatGPT for Excel, bringing GPT-5.4 directly into Microsoft spreadsheets.

Why it matters: The 1M context window changes architecture decisions for RAG pipelines and enterprise document processing. Full analysis in our GPT-5.4 review.

2. Google Drops Gemini 3.1 Flash-Lite for High-Volume Workloads

Google launched Gemini 3.1 Flash-Lite, optimized for speed and cost. Input tokens cost $0.075 per million, making it 33x cheaper than GPT-5.4 for high-volume tasks. Our benchmarks showed it handles 340 concurrent requests before latency spikes, compared to 180 for GPT-5.4.

Why it matters: Flash-Lite is now the default choice for production workloads where speed and cost outweigh reasoning depth. See our head-to-head comparison.

3. Alibaba’s Qwen 3.5 Sets New Open-Source Benchmarks

Alibaba released Qwen 3.5, a 72B mixture-of-experts multimodal model that outperforms Llama 4 Maverick and Mistral Large 3 on standard evaluations. It runs on accessible hardware thanks to its MoE architecture (14B active parameters per forward pass) and ships under the Apache 2.0 license for full commercial use.

Why it matters: The gap between open-weight and closed-source models is shrinking fast. Teams that cannot send data to third-party APIs now have a competitive alternative.

4. WHO Publishes Advisory on AI Chatbots in Mental Health

The World Health Organization issued a formal advisory warning about AI chatbots used for emotional support. The advisory cited unreliable crisis detection, data privacy risks, and the lack of clinical validation in current products. Over 20 million people use AI mental health tools globally.

Why it matters: This advisory will shape regulation of AI health apps worldwide and puts pressure on companies like Replika, Woebot, and Character.AI to add clinical safeguards.

5. Global AI Spending Reaches $2.52 Trillion

IDC’s latest forecast puts 2026 worldwide AI spending at $2.52 trillion, a 44% jump from 2025. Infrastructure (GPUs, data centers) accounts for 35% of the total. Financial services leads sector spending at $340 billion, followed by healthcare at $280 billion and manufacturing at $260 billion.

Why it matters: AI investment is accelerating, not plateauing. The data shows that 88% of companies now use AI in at least one business function.

6. NVIDIA Details Rubin GPU Platform

NVIDIA released additional specifications for its Rubin platform, including 288GB HBM4 memory per GPU and NVLink 6 at 3.6 TB/s bandwidth. The platform promises 2x training throughput over Blackwell B200. Cloud availability is expected in Q4 2026.

Why it matters: Rubin is NVIDIA’s answer to growing demand for AI training compute. The HBM4 bandwidth increase is particularly significant for LLM inference workloads.

7. EU AI Act Enforcement Phase Begins

The EU AI Act’s enforcement provisions went into effect in March 2026. Banned AI practices (social scoring, manipulative techniques, unauthorized biometric identification) are now enforceable with fines up to 7% of global revenue. High-risk system compliance deadlines arrive in August 2026.

Why it matters: This is the world’s first comprehensive AI regulation with real enforcement teeth. US companies serving European customers need to audit their AI systems now.

“This was one of the most consequential weeks for the AI industry in years. Three major model releases, a regulatory enforcement milestone, and spending data that shows the market is accelerating. The pace is not slowing down.”

What to Watch Next Week

Microsoft is expected to announce expanded Copilot Cowork features at its Enterprise AI Summit. Google has teased updates to its Workspace AI integrations. And several research labs are expected to publish results on agentic AI evaluation frameworks.

Follow our coverage throughout the week for in-depth analysis on each story.