HackAI Newsletter — May 2026

Missed our last update? Read the Early April 2026 newsletter →

🗓️ Upcoming Events

🤖 HackAI x webAI — Monday May 18th 6:00pm - 8:00pm at Capital Factory (Voltron Room) — We're teaming up with webAI, Austin's own $2.5B private AI platform, for a night of building, demos, and deep dives into on-device AI. webAI has been open-sourcing some impressive tools lately (more below) — come see them in action. Pizza, networking, and good vibes as always. RSVP here →

🙌 Sponsor Spotlight: webAI

This edition is brought to you by webAI — an Austin-based AI company (HQ'd on Congress Ave) building private, on-device AI for enterprises. Valued at $2.5B, they're on a mission to run AI on your hardware without cloud dependencies. And they're putting their money where their mouth is with their most recent open-source drop:

👁️ YOLO26-MLX — A ground-up implementation of YOLO26 real-time object detection written natively for Apple's MLX framework. No PyTorch dependency at runtime — just pure Metal GPU acceleration on Apple Silicon. It's up to 2x faster than PyTorch MPS for inference and nearly 3x faster for training. If you're building computer vision on a Mac, this is a big deal. See the repo on GitHub →

Thanks to webAI for sponsoring this edition and supporting the HackAI community! 🙏

🌐 Big AI Headlines

🚀 New: OpenAI Drops GPT-5.5 — Then Makes It Everyone's Default in Two Weeks OpenAI launched GPT-5.5 on April 23, calling it their "smartest and most intuitive" model yet, with major gains in agentic coding, computer use, and research. Two weeks later on May 5, they rolled out GPT-5.5 Instant as the new default for all ChatGPT users — boasting 52.5% fewer hallucinations and shorter, less emoji-heavy responses. The same day, OpenAI launched a self-serve Ads Manager with CPC bidding, targeting $2.5 billion in ad revenue this year. The "super app" vision is taking shape fast. Learn more →

🇨🇳 Update: DeepSeek V4 Finally Launches — Open-Source and Dangerously Cheap Last time: DeepSeek V4 was the most-anticipated and most-delayed model of 2026, missing every launch window since February. We held it from the headlines until it actually shipped. Big update: It shipped. DeepSeek dropped V4 Preview on April 24 — the same day as GPT-5.5 — with two variants: V4-Pro (1.6 trillion parameters, 49B active) and V4-Flash (284B params, 13B active). Both support 1 million token context. The price: ~$1.74 per million input tokens vs. $5 for GPT-5.5 and Claude Opus 4.7. It's open-weight, so you can run it yourself. Benchmarks put it near-parity with GPT-5.4 on math and reasoning. And Huawei confirmed its Ascend 950PR chip runs V4 at nearly 3x the inference performance of Nvidia's China-specific H20. Still in preview, but the AI cost curve just got steeper. Learn more →

🤝 New: Anthropic Signs Musk's SpaceX for Compute — Revenue Up 80x In a move nobody saw coming three months ago, Anthropic signed a deal to use all of SpaceX's Colossus 1 data center capacity — over 220,000 Nvidia GPUs and 300+ megawatts of compute. CEO Dario Amodei said revenue and usage grew 80-fold in Q1 2026 (annualized revenue now ~$30B), when they'd only planned for 10x. "That is the reason we have had difficulties with compute," Amodei said. Musk went from calling Anthropic "evil" in February to saying "no one set off my evil detector" after meeting the team. Oh, and xAI was dissolved as a separate company — it's now SpaceXAI. Strange bedfellows make the best compute deals. Learn more →

⚔️ Update: Pentagon Signs AI Deals with 7 Rivals — Anthropic Still Frozen Out Last time: Anthropic won a preliminary injunction in San Francisco, with Judge Rita Lin calling the Pentagon's actions "classic illegal First Amendment retaliation." Now: The DC appeals court denied Anthropic's separate bid to lift the supply chain risk label on April 8. Then on May 1, the Pentagon signed deals with SpaceX, OpenAI, Google, Nvidia, Microsoft, AWS, Oracle, and Reflection for classified AI work — pointedly excluding Anthropic. But the door isn't shut: Dario Amodei visited the White House on April 17, Trump called Anthropic "high-IQ" and said a deal was "possible," and Pentagon insiders told Reuters they're "reluctant to give up" Claude. This saga has more twists than a Netflix series. Learn more →

⚖️ New: Musk v. Altman Goes to Trial — And the Revelations Are Wild The biggest AI trial of the decade kicked off April 28 in Oakland. Elon Musk is suing Sam Altman and Greg Brockman, alleging they "stole a charity" by converting OpenAI from nonprofit to an $852B for-profit. The fireworks so far: Musk admitted under oath that xAI uses OpenAI's models for training (audible gasps in the courtroom). Brockman's own 2017 journal surfaced with entries about being "warm to steal the nonprofit." And Shivon Zilis — former OpenAI board member and mother of four of Musk's children — testified that Musk offered Altman a Tesla board seat. A verdict could derail OpenAI's planned ~$1T IPO. The trial continues. Learn more →

🏛️ Update: White House Considers Pre-Release AI Vetting — A Major Policy Reversal Last time: We covered Project Glasswing and debated whether frontier AI companies should build offensive cyber capabilities at all. Now: Glasswing's Mythos model is reshaping U.S. AI policy in real time. The White House is drafting an executive order to vet AI models before public release — comparing the process to FDA drug approval. NEC Director Kevin Hassett said they've "scrambled an all of government effort" to coordinate. Google, Microsoft, and xAI have agreed to give the government early model access. This is a stunning reversal for an administration that scrapped Biden's AI oversight rules on Day 1. Mythos changed the calculus. Learn more →

🤠 Texas AI Spotlight

🏭 Update: Terafab Gets Intel 14A — And a $55 Billion+ Price Tag Last time: Intel announced it was joining the Terafab project, adding serious silicon muscle to the largest AI chip facility announcement in Texas history. Big update: During Tesla's Q1 earnings call on April 23, Musk confirmed Terafab will use Intel's next-generation 14A process — making Tesla the first major external customer for the technology. Tesla is investing $3 billion in a semiconductor R&D facility at GigaTexas for pilot production, while SpaceX handles high-volume manufacturing. A public notice filed in May puts SpaceX's initial investment at $55 billion, with a total potential cost up to $119 billion across all phases. For context, Samsung's Taylor fab was $17B. This is another scale entirely. Learn more →

⚡ Update: Gas Leapfrogs Wind in Texas Grid Queue — First Time in a Decade Last time: ERCOT flagged 24 gigawatts of new data center demand by 2031, with oil majors building gas plants specifically for West Texas data centers. Now: The data center boom has officially reshuffled Texas energy. Gas generation has overtaken wind in ERCOT's interconnection queue for the first time since January 2016. The numbers are staggering: 360,000 megawatts of data center demand in the queue — enough to quadruple the grid's all-time peak. ERCOT CEO Pablo Vegas told the Senate that total incoming demand is 410,000 MW, with 87% from data centers. Gas projects have jumped 400% in three years. The AI boom is literally rewriting how Texas generates electricity. Learn more →

🏛️ New: Texas House Digs Into Data Center Impacts The Texas House Committee on State Affairs held hearings with data center developers, operators, and ERCOT leadership. CEO Pablo Vegas revealed ERCOT is building an entirely new batch interconnection process to manage unprecedented demand — a major shift from Texas's hands-off approach to grid planning. Much of the current queue is speculative: developers with land and power lines filing requests with no real projects behind them. Meanwhile, Maine became the first state to pass a data center moratorium — but that idea hasn't surfaced in Austin. Yet. Learn more →

🌬️ New: Texas Approves First Data Center + Wind Farm Direct Power Deal In a unanimous 5-0 vote on April 24, the Public Utility Commission of Texas approved a first-of-its-kind arrangement: a data center drawing electricity directly from an existing wind farm, bypassing traditional grid interconnection. It's a creative workaround for the years-long queue bottleneck. If successful, it could become a template for pairing renewable energy with large-scale AI compute across the state. Critics worry it shifts costs onto other ratepayers — but with the queue growing by the month, expect more creative solutions like this. Learn more →

🗣️ This Month's AI Debate

Should the government vet AI models before release — like the FDA does with drugs?

The Trump administration is now considering a pre-release review process for frontier AI models. The trigger? Anthropic's Mythos, which can find zero-day vulnerabilities faster than any human red team. NEC Director Kevin Hassett compared it to FDA drug approval: "so that they're released to the wild after they've been proven safe."

Last month we debated whether AI companies should build offensive cyber capabilities. This month, the question moves upstream: who decides when a model is safe enough to release?

🏛️ Pro Vet It: Mythos found a 27-year-old OpenBSD bug in hours. If one model can do that today, future models will do more. Without a review process, the first time we learn a model has dangerous capabilities is after those capabilities are deployed. Google, Microsoft, and xAI have already agreed to provide early access. A structured vetting process — with clear timelines and industry participation — protects both innovators and the public.

🚀 Con Don't Gate It: FDA drug trials take 10+ years. AI models ship in weeks. A government review bottleneck could hand China a decisive lead while American models sit in bureaucratic limbo. The voluntary testing framework at the Commerce Department is the better path — structure without strangulation. And if the government couldn't secure its own networks against decades-old bugs, is it really the right body to evaluate cutting-edge AI?

💬 What do you think? Bring your hot takes to the next HackAI Meetup!

Community CTA

Got an AI project, job posting, or demo to share? Reply to this email or message us on Luma — we'd love to feature you!

Stay curious, Austin 🤖💡 — The HackAI Team, Reid & Bryce