Qualcomm Buys Modular for $3.9B, OpenAI Unveils First AI Chip — June 24 AI Roundup
Qualcomm Buys Modular for $3.9B, OpenAI Unveils First AI Chip — June 24 AI Roundup
June 24, 2026 — Another packed day in AI. Here's everything you need to know.
🏆 Top Story: Qualcomm Acquires Modular for $3.9 Billion
Qualcomm announced it will acquire AI infrastructure startup Modular in a $3.9 billion all-stock deal, marking the chipmaker's largest AI bet yet.
Modular is best known for building MAX, an alternative to the CUDA ecosystem that lets AI models run across any hardware (NVIDIA, AMD, Intel, or Qualcomm itself) without rewriting code. Their flagship product is MAX Engine, an inference engine that claimed 3-10x speed improvements over TensorFlow and PyTorch on commodity hardware.
Why it matters: Qualcomm has been trying to break into the AI data center business but has been held back by software — NVIDIA's CUDA moat is the deepest in the industry. With Modular's software stack, Qualcomm gains a CUDA-compatible runtime that could let its chips run existing AI workloads without modification. The deal also bolsters Qualcomm's on-device AI strategy — Modular's compiler technology works for mobile and edge as well as data centers.
Market context: The acquisition comes amid growing industry desire to reduce dependence on NVIDIA. AMD, Intel, and a wave of startups are building competing hardware, but all share the same problem: CUDA lock-in. Modular's value proposition is that it breaks that lock-in — and Qualcomm just bought the key.
🔥 OpenAI + Broadcom Unveil First Custom AI Chip
OpenAI and Broadcom jointly announced their first custom AI chip, designed to reduce OpenAI's dependence on NVIDIA GPUs and slash inference costs.
The chip is purpose-built for transformer model inference — the architecture behind GPT, Claude, Gemini, and every major LLM. Early benchmarks suggest the chip delivers comparable performance to NVIDIA H200 GPUs for inference workloads while consuming significantly less power. OpenAI expects the chip to cut inference costs by 30-50% once fully deployed.
What this means for the industry:
- OpenAI joins Google (TPU), Amazon (Trainium), Microsoft (Maia), and Meta in the custom chip club
- Broadcom gains a marquee AI customer in the design services business
- NVIDIA's data center dominance faces another structural challenge
- Lower inference costs = cheaper API pricing = faster AI adoption
🇨🇳 China Blacklists 56 US Firms — AI Export War Goes Bidirectional
China's Ministry of Commerce retaliated against the Pentagon's expanded 1260H list by blacklisting 56 American companies:
- 46 firms banned from Chinese government procurement
- 10 firms added to dual-use export control list
Targeted sectors: Drones (Teal Drones, Jaia Robotics), aerospace (Ball Aerospace), maritime defense (L3Harris), and critically — rare earth mining. MP Materials and USA Rare Earth are on the list. China controls ~70% of global rare earth mining and ~90% of refining capacity.
The context: June 12's Fable 5 export control was the first time the US applied controls to a deployed commercial AI API. China's response targets the physical supply chain for chips and electronics. The AI export war is no longer one-directional — it's a two-front conflict (software vs. hardware, AI models vs. raw materials).
📸 Getty-OpenAI Deal Sends GETY Stock +145%
Getty Images signed a multi-year display partnership with OpenAI to bring licensed editorial photography into ChatGPT search and discovery.
- Not a training deal — display rights only, same model as Getty's Perplexity deal
- Users will see real licensed photos in ChatGPT for news, travel, history, and product queries
- GETY surged 145% on the announcement
- Getty was previously the most-shorted stock in the market — the deal transforms the narrative from "AI victim" to "AI infrastructure provider"
📊 More Headlines
| Headline | Impact |
|---|---|
| SK Hynix files $29B US IPO — all proceeds go to AI memory fabs | Signal: AI memory demand is exploding |
| GPT-5.6 rumors — community reports silent upgrade in ChatGPT/Codex | Unconfirmed, but pattern matches past silent releases |
| OpenAI targets $100B ad revenue by 2030 — ad layer for ChatGPT | First major public signal: ChatGPT as an ad platform |
| AI stocks rebound after global selloff — "no cracks in the armor" | Short-term volatility, long-term thesis intact |
| Meta pauses employee keystroke tracker for AI training | Privacy pushback at scale |
| MGX sets up $50B AI fund | Abu Dhabi's sovereign wealth goes all-in on AI |
| US government urges Meta to share AI models | Policy push for open-weight model access |
| Qualcomm also buys Edge Impulse — edge AI platform | Edge + data center = full stack |
🔮 Takeaway
Today's biggest theme is vertical integration. Qualcomm buying Modular for the software stack. OpenAI building its own chip with Broadcom. SK Hynix going public to fund fab expansion. Even Getty Images repositioning from AI victim to AI partner.
The AI industry is maturing from a single-vendor (NVIDIA) world into a multi-layered ecosystem where every layer is being vertically re-integrated — chips, software, models, content, and infrastructure. The next 12 months will define which players own which layers.
Catch up on previous roundups: AI News Archives
Stay updated with the latest AI tools and news at Mee AI.
Found this helpful? Share it with your team.
Read more articles →