Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
DeepSeek released V4-Pro (1.6T parameters, 49B active) and V4-Flash (284B total, 13B active) as open-source previews with 1M token context. Benchmarks rival GPT-5.5 and Claude Opus 4.7, but Flash pricing starts at $0.14 per million input tokens — roughly 36x cheaper than GPT-5.5. Both models are available on Hugging Face and via DeepSeek's API, reigniting the US-China AI cost-performance race.
OpenAI's newest flagship went live April 23-24 for paid subscribers and API users. GPT-5.5 uses fewer tokens to complete the same tasks as its predecessor while matching GPT-5.4's speed, and is positioned as a unified agent that writes code, browses the web, operates software, and chains tools autonomously. API pricing: $5/$30 per million tokens (input/output); Pro variant at $30/$180.
Google's marquee conference delivered a full agentic stack: the Gemini Enterprise Agent Platform with low-code Agent Studio, no-code Agent Designer, and centralized Agent Inbox. Ironwood (7th-gen TPU) is now GA for cloud customers. Google previewed TPU 8 as two purpose-built chips — 8t for training (with Broadcom, 2.8x over Ironwood) and 8i for inference (with MediaTek, 20-30% cheaper) — both at TSMC 2nm targeting late 2027. Anthropic's Claude Opus 4.7 was added to the platform.
The Information reported that Brin is directly involved in a new DeepMind unit led by Sebastian Borgeaud, focused on long-horizon agentic coding — exactly where Claude Code dominates. In an internal memo, Brin wrote: "To win the final sprint, we must urgently bridge the gap in agentic execution and turn our models into primary developers." Google says AI writes about 50% of its code; Anthropic claims near-100%.
Meta will deploy tens of millions of AWS Graviton5 cores (3nm) in a deal lasting at least three years, making it one of the largest Graviton customers worldwide. The deal signals a structural shift: while GPUs dominate training, agentic AI workloads — real-time reasoning, code generation, multi-step orchestration — are driving massive CPU demand. This follows $48B in recent AI infrastructure commitments from Meta.
Anthropic's expanded Amazon partnership brings total investment to $13B at a $350B valuation, with commitments for up to 5GW of compute capacity including Trainium2 and Trainium3 chips. Anthropic's annualized revenue has surpassed $30B (up from ~$9B at end of 2025), with over 100,000 customers running Claude on Amazon Bedrock.
Apoorva Mehta's new fund uses thousands of AI agents to autonomously identify trades, research stocks, size positions, and execute — essentially replacing fundamental portfolio managers. Abundance raised $100M in seed and currently trades its own capital. Mehta says he was inspired by OpenAI's o3 model demonstrating complex autonomous reasoning.
Tesla's 2026 spending plan has ballooned to $25B — up from $8.5B last year — targeting AI compute, chip design, a semiconductor fab in Austin, and robotaxi expansion. Fully unsupervised robotaxi rides launched in Dallas and Houston on April 18, with Phoenix, Miami, and Las Vegas planned for H1. The company expects negative free cash flow for the rest of 2026.
The former OpenAI executive's startup signed a new multibillion-dollar agreement to expand its use of Google Cloud AI infrastructure, following an earlier partnership. The deal underscores Google's strategy of locking in high-profile AI labs as cloud customers, even those that may compete with its own models.
Three AI bills remain live as Arizona's legislature adjourns today: HB 2592 (mandating state agencies adopt AI to cut red tape), HB 2311 (children's chatbot safety), and SB 1786 (broader AI regulation headed back to the Senate for concurrence). Arizona could become the third state after Utah and Washington to pass AI legislation this session.
