AI News: May 5, 2026

1. Cerebras Files Updated IPO Terms: $3.5B Raise, Up to $26.6B Valuation

Cerebras. The wafer-scale AI-chip maker set a price range of $115–$125 per share for 28 million shares, targeting roughly $3.5 billion in proceeds and a fully-diluted valuation up to $26.6 billion when it lists on Nasdaq as CBRS. Cerebras reported Q4 revenue of $510M (+76% YoY) and $87.9M in net income, and is leaning on a $20B+ multi-year compute deal with OpenAI for up to 750 MW through 2028. Bankers expect pricing the night of May 13 with trading May 14, in what will be the most direct public-market read on whether anyone other than NVIDIA can own the AI training market. Source

2. Sierra Raises $950M Series E at a $15.8B Valuation

Sierra. Bret Taylor’s customer-experience agent company closed a $950M Series E led by Tiger Global and GV (with Benchmark, Sequoia, and Greenoaks following), pushing post-money to $15.8B and giving Sierra over $1B on the balance sheet. The company says it has crossed $150M ARR in eight quarters and now serves more than 40% of the Fortune 500, including Prudential, Cigna, Blue Cross Blue Shield, and Rocket Mortgage. The round is the clearest signal yet that VCs see AI customer service as a genuinely large, defensible category — not a feature inside Salesforce. Source

3. Banks Are Quietly Underwriting the AI-Data-Center Buildout

The Decoder. A reported piece argues that the financing structures behind hyperscaler AI capex — long-dated debt against power-constrained sites with single-tenant offtake risk — are becoming a meaningful concentration risk for the banks underwriting them. With multi-year compute deals (OpenAI/Oracle, OpenAI/Cerebras, Anthropic/Google) increasingly used as collateral, lenders are taking on exposure that looks more like utility project finance than tech lending. Worth reading if you’ve been wondering what happens if frontier-model demand cools mid-buildout. Source

4. DoorDash Ships AI Onboarding, Dish-Photo Editor, and Merchant Site Generator

DoorDash. DoorDash rolled out a set of AI tools for restaurants: faster menu and merchant onboarding, photo editing for dish images, and an AI website generator for merchants without an existing site. The features target the long tail of independent restaurants where missing menu data and weak photography measurably depress conversion — a small-but-real example of AI being deployed where the marginal value is high precisely because the prior baseline was so low. Source

5. Image AI Has Quietly Overtaken Chatbots as the App-Store Growth Driver

TechCrunch. New data shows image-generation apps are now driving more downloads than chatbot upgrades, though they continue to underperform chatbots on revenue per user. The pattern is consistent across both stores: image is the consumer wedge, conversation is where the money is — at least until chat experiences get cheaper to deliver and image experiences find a recurring use case. Source

6. EU AI Act Compliance Deadline Wobbles as Trilogue Stalls

Holland & Knight / EU. The April 28 trilogue between the European Parliament, Council, and Commission on the Digital Omnibus failed to reach agreement, with another meeting scheduled for May 13. If the Omnibus passes as drafted, the August 2, 2026 compliance deadline for high-risk employment-related AI systems would slide to December 2, 2027. U.S. companies serving EU users should plan for August either way: each member state still must stand up at least one AI regulatory sandbox by that date under Article 57. Source

7. Import AI 455: The Case That Self-Improving AI Is Already Underway

Import AI. Jack Clark’s latest issue makes the case — citing benchmark progress on coding, scientific reproducibility, and alignment-research automation — that there’s a >60% probability of fully automated AI R&D (AI systems autonomously training their successors) by end of 2028. The piece reads less like a hot take and more like a clinical inventory of which pieces are already in place; it’s worth the read whether or not you buy the timeline. Source