Nvidia can't make chips fast enough for China
Here's the one thing to take from today: the bottleneck in AI is still chips and power, and both are getting harder to buy. Nvidia told partners it could fill only 5% to 10% of some requests, and the big UK data campus won't start operating until early 2028. If you build on top of this, plan for scarce, expensive compute and look for tools that let a small team do more with less.
- Nvidia chips stuck in China demand
- Brookfield and Drax plan a $27B campus
- Testifly runs your QA without a hire
- Quick hits: agents, robots and models
- AI Index slid 19.5% over 30 days
Top stories
All news →What happened: Alibaba, Tencent, ByteDance and others have reportedly ordered about 6 million Nvidia workstation chips, but expect only a small slice soon.
- Bloomberg reported on November 4, 2026, that the buyers ordered about 6 million RTX Pro 5500 cards, worth roughly $70 billion at more than $11,000 each.
- Buyers expect only some tens of thousands of units within four months, about 1% of the total. Investing.com gave about ten thousand, but 1% of 6 million is closer to 60,000.
- Reuters reported on October 20 that ByteDance, Alibaba and Tencent alone ordered more than 3 million cards at about $11,400 each. Nvidia has not confirmed the totals.
Why it matters: The RTX Pro 5500 sits outside the export limits on Nvidia's most advanced chips, so it is one of the strongest parts Chinese firms can still buy. For founders, expect hardware supply to stay tight and prices to stay high, and build your compute budget with that in mind.
What happened: Drax, Brookfield and Radiant plan an AI data campus on the Drax Power Station estate that could eventually pass 1 gigawatt of computing capacity.
- Phase one is 100MW, with £2.5 billion (about $3.4 billion) of initial investment and operations targeted for early 2028.
- Brookfield estimates the full build, including semiconductors, could exceed £20 billion ($26.8 billion), with a further 500MW planned between 2028 and 2031.
- Phase one cannot proceed until planning, regulatory and network approval come through. Drax shares rose more than 8% on the announcement.
Why it matters: Land, power and grid access are becoming the scarce inputs for AI, not just chips. Capacity like this is years away, so founders should plan where their future compute will come from and what it will cost.
What happened: Testifly is an autonomous QA tool built for teams of one to five people that cannot justify hiring a QA engineer.
- You submit your app's URL, and Testifly finds real user flows, builds tests, runs them on every deploy and on a schedule, and repairs them when the interface changes.
- It writes tests in four groups: visibility, functionality, business logic and negative testing, and it connects to Jira, Linear, Notion and Slack. Failed tests come with a video explaining the failure.
- There is a 30-day free trial with self-serve signup, no demo call or procurement step, and the launch drew 252 upvotes on Product Hunt.
Why it matters: Teams too small to hire a QA engineer often ship with thin testing. This gives them a way to test every release without adding a headcount, so ask whether that hire is still on your plan.
Quick hits
- Scalekit's AgentKit gives agents 30,000+ tools across 600+ apps through one gateway, free up to 10,000 calls a month. Product Hunt →
- Synter's AI agent runs ads across 30+ platforms, charging $0.25 per verified write and 0% of ad spend. Product Hunt →
- Teamily AI puts people and always-on AI agents in one messenger, free to start, with 342 Product Hunt upvotes. Product Hunt →
- Neufangled's open-source Stringman, listed at $795, hangs from ceiling anchors and moves objects around a room. neufangled.com →
- Qwen running on a laptop can make sense of model weights, per simonwillison.net. Full brief coming. Read →


