#1DeepSeek V4 Launch Imminent — and It Runs on Huawei Chips
DeepSeek's V4, a 1-trillion-parameter multimodal MoE model, is expected to drop in the final weeks of April — and Reuters confirmed it will run exclusively on Huawei's Ascend 950PR chips. DeepSeek deliberately locked out NVIDIA, handing Chinese chipmakers a landmark first: a frontier model built entirely on domestic semiconductor infrastructure. Alibaba, ByteDance, and Tencent have already placed bulk orders for Huawei Ascend hardware.
Source: [FindSkill.ai](https://findskill.ai/blog/deepseek-v4-release-date-specs/) | [Dataconomy](https://dataconomy.com/2026/03/16/deepseek-v4-and-tencents-new-hunyuan-model-to-launch-in-april/)
---
#2Google's TurboQuant Cuts LLM Memory 6x — No Retraining Required
Google Research's TurboQuant algorithm compresses LLM key-value caches to 3 bits, delivering a 6x memory reduction and 8x speedup in attention computation on H100 hardware — with no accuracy loss and no model-specific tuning needed. It's architecture-agnostic, slotting into any transformer, and integration into major serving frameworks like vLLM and TensorRT-LLM is expected later this year. This quietly reshapes the inference cost curve for on-device and cloud AI alike.
Source: [Digitimes](https://www.digitimes.com/news/a20260327VL207/google-llm-ai-inference-cost-algorithm.html) | [Google Research](https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/)
---
#3Google Gemma 4 Drops as Apache 2.0 — A Big Open-Source Signal
Google released Gemma 4 on April 2nd under the Apache 2.0 license, removing the usage restrictions that had hampered earlier Gemma models. The license change is the headline: it means developers and companies can now build commercial products on top of Gemma without legal friction. This is a direct competitive move against Meta's Llama ecosystem.
Source: [Agentic AI Hype Substack](https://agenticaihype.substack.com/p/anthropics-500k-line-code-leak-google)
---
#4Anthropic's MCP Crosses 97 Million Installs — Agentic Standard Taking Shape
Anthropic's Model Context Protocol hit 97 million installs in March 2026, signaling its shift from experimental spec to foundational infrastructure for AI agents. Every major AI provider now ships MCP-compatible tooling, and the Linux Foundation's Agentic AI Foundation — anchored by MCP, OpenAI's AGENTS.md, and Block's goose — is formalizing the governance layer around it.
Source: [LLM Stats](https://llm-stats.com/llm-updates) | [Crunchbase](https://news.crunchbase.com/venture/foundational-ai-startup-funding-doubled-openai-anthropic-xai-q1-2026/)
---
#5OpenAI Hits $2B/Month in Revenue, Eyes IPO
OpenAI is now generating $2 billion in monthly revenue and is actively preparing for a public offering, riding momentum from its $122 billion Q1 raise at an $852 billion valuation. The revenue milestone represents a sharp acceleration — it took the company years to reach $1B annually, and now that figure passes in weeks.
Source: [AI & News](https://www.aiandnews.com/blog/latest-ai-news-april-2026/) | [Crunchbase](https://news.crunchbase.com/venture/foundational-ai-startup-funding-doubled-openai-anthropic-xai-q1-2026/)
---
#6Ledger CTO: AI Is Driving Crypto Exploit Costs "Down to Zero"
Ledger's CTO Charles Guillemet warned that AI has made reverse-engineering software and chaining exploits a matter of seconds rather than months, and $1.4 billion in crypto losses over the past year may be a floor, not a ceiling. AI-powered deepfakes now impersonate customer support and influencers at scale, while automated tools scan compromised phones for seed phrases without any user interaction.
Source: [CoinDesk](https://www.coindesk.com/tech/2026/04/05/ai-is-making-crypto-s-security-problem-even-worse-ledger-cto-warns/)
---
#7Q1 2026 AI Venture Funding Doubles All of 2025
Foundational AI startup funding in Q1 2026 hit $297 billion — more than double the entirety of 2025 — driven by OpenAI's $122B raise, Anthropic's $30B Series G, and xAI's $20B Series E. The capital concentration at the frontier layer is accelerating, raising real questions about whether the money is flowing toward capability or toward moat-building.
Source: [Crunchbase](https://news.crunchbase.com/venture/foundational-ai-startup-funding-doubled-openai-anthropic-xai-q1-2026/)
---
**Big Picture**
Today's theme is fracture and consolidation happening simultaneously. On the fracture side: DeepSeek's deliberate Huawei-first strategy signals a maturing AI cold war where chip access and model access are now inseparable policy levers. On the consolidation side: Google's TurboQuant, Gemma 4 under Apache, and Anthropic's MCP milestone all point toward infrastructure standardization — the boring, load-bearing work that determines who controls the agentic layer once the model wars stabilize. Meanwhile, the Q1 funding numbers confirm that capital has already picked its winners. The real race now is execution.
