AI News: April 19, 2026

1. Cerebras Refiles for IPO with $510M Revenue and AWS, OpenAI Deals on the Books

Cerebras. The wafer-scale AI chip startup filed for a mid-May IPO on its second attempt, disclosing $510M in 2025 revenue and $237.8M GAAP net income. The S-1 reveals an AWS infrastructure partnership, a reported $10B+ inference deal with OpenAI, and a $23B valuation from February’s $1B Series H. The listing will be the first serious public-market test of whether inference-specialist silicon can sustain NVIDIA-adjacent multiples outside the hyperscaler bubble. Source

2. Recursive Superintelligence Raises $500M at $4B Valuation With Nothing Shipped

Recursive Superintelligence. Richard Socher (ex-Salesforce chief scientist) and Tim Rocktäschel (ex-DeepMind), plus about 18 researchers from OpenAI, Google, and Meta, closed at least $500M — with reports suggesting up to $1B — at a $4B pre-money valuation led by GV and NVIDIA. The pitch is “recursive self-improvement” as the path to superintelligence, with no product or public research yet. It is a blunt data point on how much capital is still chasing labs whose only artifact is a thesis. Source

3. DeepSeek Seeks First-Ever Outside Funding at $10B Valuation as V4 Slips

DeepSeek. The Chinese open-weight lab is reportedly in talks for at least $300M at a $10B+ valuation — its first external raise after being solely funded by High-Flyer Capital Management. V4 has slipped as engineers retool training for Huawei Ascend chips under Beijing’s semiconductor self-sufficiency push, and key researchers have reportedly defected to Xiaomi and ByteDance. The raise signals that even capital-rich Chinese labs need outside money to compete, and that Huawei-hosted training is becoming the default for China’s frontier stack. Source

4. Tesla Robotaxi Expands to Dallas and Houston

Tesla. Tesla launched driverless robotaxi service in Dallas and Houston, bringing its Texas footprint to three cities plus limited Bay Area operations still using human drivers. Initial deployment is tiny — one registered vehicle per new city versus 46 in Austin — and Austin’s fleet has logged 14 crashes since launch per a February regulatory filing. The expansion narrows the operational gap with Waymo (500K+ paid rides/week across 10 cities) while the safety data keeps the underlying question open. Source

5. App Store Launches Jump 104% Year-Over-Year as AI Coding Lowers the Barrier

App Store. New data shows Q1 2026 app releases up roughly 60% year over year across both major stores (80% on iOS), with April running at +104% year over year combined and +89% on iOS. The piece credits AI-assisted coding tools like Claude Code and Replit for lowering the barrier for non-technical creators, and flags that Apple’s review capacity is straining — a fake Ledger Live clone recently drained $9.5M in crypto. Empirical evidence that “AI will kill apps” was premature; the harder question is how review systems scale to match shipping velocity. Source

6. Study: Ten Minutes of AI as an Answer Machine Measurably Erodes Problem-Solving

Liu et al., arXiv:2604.04721. A US and UK team ran three controlled experiments on fraction problems and SAT reading comprehension. Participants who used GPT-5 for direct answers performed significantly worse and skipped more problems after AI was removed; the 61% who used AI for direct answers showed the steepest declines, while users who ignored AI entirely scored best on unassisted tests. Implication for practitioners: design AI interfaces to scaffold reasoning rather than hand over answers, and revisit how junior engineers are onboarded under heavy AI assistance. Source