Good morning. It's Friday, April 24th, 2026.

**China's DeepSeek ships V4.** After months of delays, DeepSeek has finally released DeepSeek-V4, its long-awaited flagship model. There are two variants: V4-Pro with 1.6 trillion total parameters and 49 billion active, and V4-Flash — a leaner 284 billion total, 13 billion active for efficient inference. Both support one million token context windows, a first for open-source. V4-Pro reportedly outperforms Google's Gemini 3.1 Pro on long-text processing, though it still trails Anthropic's Claude Opus 4.6. But the big news isn't the benchmarks — it's the hardware. DeepSeek and Huawei worked closely to optimize V4 for Huawei's Ascend chips, marking a significant pivot from the company's previous reliance on Nvidia hardware. Jensen Huang called this a "horrible outcome for our nation." Chinese chipmakers rallied on the news, with SMIC up ten percent. This is the decoupling playing out in real time.

**OpenAI fires back with GPT-5.5.** Less than two months after GPT-5.4, OpenAI launched GPT-5.5 on Thursday, rolling it out now to all paid ChatGPT and Codex subscribers. OpenAI President Greg Brockman says the key advance is that GPT-5.5 "can look at an unclear problem and figure out just what needs to happen next" with far less guidance. It excels at coding, debugging, data analysis, and deep research — in other words, agentic work. OpenAI also disclosed that GPT-5.5 meets its "High" cybersecurity risk classification, though it stays below the "Critical" threshold. The API is coming "very soon," pending additional safeguards. The cadence is striking — six weeks between major model versions. That's not iterative improvement. That's an arms race.

**Tesla's Terafab stakes a claim on Intel 14A.** During Tesla's Q1 earnings call, Elon Musk confirmed that the Terafab project — a joint AI chip manufacturing venture with SpaceX, Tesla, and xAI — will use Intel's next-generation 14A process technology. Intel joined the Terafab project in early April, bringing established chipmaking expertise to what Musk envisions as a vertically integrated silicon complex. Tesla also raised its 2026 capital spending to over $25 billion, mostly directed at AI and robotics. For Intel, this is a potential lifeline for its foundry ambitions after years of lagging TSMC. For the rest of the industry, it signals that the biggest chip consumers are starting to make their own.

**Artemis II heat shield verdict is in.** NASA confirmed this week that the Orion heat shield performed as expected during Artemis II's re-entry, with no unusual conditions identified. Navy diver imagery taken after splashdown on April 10th showed the char loss issues that plagued Artemis I were significantly reduced — both in quantity and size. The fix was simple but controversial: NASA dropped the skip re-entry maneuver entirely. Former astronaut Charles Camarda had called the plan to fly with the shield "a very bad decision." Turns out, the trajectory change was enough. The shield now goes to Marshall Space Flight Center for detailed scans. Meanwhile, NASA has already rolled out the Artemis III core stage. The program is moving.

That's all for today. Stay curious.