Trending Thursday #28

In partnership with

Hi !

Here's what's been trending in Tech & VC this week:

• The Power Crunch Reality Check: Big Tech's AI data center ambitions face a physics problem that could "deflate the bubble" before it fully inflates

• The $60B Global Land Grab: Microsoft and Amazon commit tens of billions to AI infrastructure in India and Canada, racing to lock up compute capacity

• Trump's 25% Chip Tax: Nvidia gets the green light to sell H200s to China, but only if the US government gets a quarter of the revenue

• The Smuggling Paradox: While Trump loosens export controls, DeepSeek trains on thousands of Blackwell chips smuggled through third countries

• Agentic AI's Constitution: Anthropic donates MCP to the Linux Foundation as 10,000+ servers go live and Big Tech unites on agent standards

As we continue to build at Context Ventures, we've launched Startup Intros with a simple mission: to help early-stage founders find the right investors, faster and smarter.

🔥 Context VC & Startup Intros Events Coming Up

Check it out, stay plugged in, and don’t miss these hot upcoming events:

• Dec 16: Verza's Annual Holiday Mixer & Content Sprint: Join a great group of top-tier creators, founders, and industry leaders. Sign up on Partiful!

• Dec 17: Startup Grind x ThinkTank VC Holiday Party: Join us for an evening of relaxed networking, great convos, and holiday cheer. Sign up here!

• Jan 20: AI Startup Pitch Night: Apps to pitch are open! Join us in SF for an exciting pitch night featuring the hottest AI startups. Sign to pitch here!

• May 27: Early Stage Summit: Due to various logistical factors, we had to reschedule the event but plan to make it better than ever. Register on Luma!

🏭 AI Infrastructure & Compute Race: When Physics Meets Ambition

⚡ The Power Crunch That Could Pop the Bubble

Big Tech's AI data center plans face a fundamental physics problem that threatens to deflate the AI "bubble" before it fully inflates:

• 44GW Need by 2028: Estimated extra capacity required for planned AI data centers

• 3-5 Year Build Time: New power gen facilities take years, but the demand is now

• Grid Competition: Data center devs are fighting for the same limited connections

• No Quick Fix: Unlike software, you can't iterate your way out of thermodynamics

The Brutal Math: Every megawatt allocated to AI training is a megawatt unavailable elsewhere. When the world's richest companies are all fighting for the same scarce resource, price isn't the constraint anymore; availability is.

Translation: Silicon Valley spent a decade telling us software would eat the world, but now the world is eating back, one kilowatt at a time.

🌏 The $60B Global Data Center Stampede

Microsoft and Amazon aren't waiting for the power problem to solve itself. They're racing to plant infrastructure flags wherever electricity and political willingness still exist:

Microsoft's Multi-Continent Play:

• $17.5B in India: Satya Nadella announced Microsoft's largest Asia investment ever after meeting with Prime Minister Modi, focused on AI and cloud computing over four years

• CA$7.5B in Canada: New AI infrastructure coming online in 2026, bringing Microsoft's total Canadian investment to CA$19B from 2023 to 2027

Amazon's Counter-Offensive: $35B+ in India by 2030: Adding to the nearly $40B Amazon has already invested, with plans to create 1M additional jobs in the country.

The Energy Play: NextEra Energy + Google Cloud: Expanding data center campuses across the US while Meta secures 2.5GW+ in clean energy contracts from the partnership.

Why India and Canada? Both have something Silicon Valley is running out of: available power capacity, land that isn't already claimed, and governments eager to host the AI future. When your home market can't deliver the electrons you need, you go where the grid can handle you.

🛰️ When Earth Isn't Enough: AI Goes Orbital

The most audacious solution to the power crunch isn't building more data centers on Earth; it's leaving Earth entirely:

Blue Origin and SpaceX Race to Space:

• Blue Origin: Has worked for over a year on tech for orbital AI data centers

• SpaceX: Plans to use upgraded Starlink satellites for AI computing payloads

• The Logic: No grid constraints, unlimited cooling (space is -270°C), no zoning battles

Starcloud's Proof of Concept:

• Already Running: Their satellite with an Nvidia H100 chip is actively querying responses from Google's Gemma

• The First: Functional AI inference happening in orbit, not just a press release

The Absurdity: We've gone from "the cloud" as a marketing metaphor to "the cloud" as a literal location. When data center constraints on Earth become severe enough, the solution is to spend $2,700 per kilogram to launch hardware into space.

Reality Check: Orbital data centers sound insane until you realize that finding 44GW of new power capacity on Earth might be even harder. At least space has unlimited solar exposure.

🇨🇳 China: Power Grids, Chip Deals & Smuggling Networks

⚡ China's Grid Advantage

China's power grid, the world's largest, gives it a structural edge in the global AI contest, helping Chinese companies develop AI models more cheaply than US competitors:

• World's Largest Grid: More interconnected capacity than any other nation

• Lower Training Costs: Cheaper electricity translates directly to cheaper AI development

• Years of Investment: Infrastructure decisions from a decade ago now pay AI dividends

The Math: AI training is essentially converting electricity into intelligence. When your power costs less and your grid can deliver more of it reliably, every training run has a cost advantage. China's infrastructure investments weren't made with AI in mind, but they're paying off anyway.

🎰 The Nvidia Export Circus: A Play in Three Acts

Act I: Trump Says Yes (With a Cut)

President Trump announced the US will let Nvidia ship H200 chips to "approved customers" in China, but with conditions that read like a mob protection arrangement:

• 25% US Government Cut: America gets a quarter of every chip sale to China

• "Approved Customers" Only: Vague criteria that create uncertainty for buyers

• National Security Review: Chips will travel to the US for inspection before export to China

• Huawei Comparison: Trump allowed exports after concluding H200s carry a lower security risk than Huawei alternatives

The Policy Reversal: David Sanger at the NYT notes this reverses decades of US tech restrictions on adversaries. Trump also said Commerce is working on the "same approach" for AMD and Intel.

Act II: Beijing Says Maybe Not

Beijing is set to limit access to Nvidia H200s despite Trump allowing their export:

• Self-Sufficiency Push: China wants to achieve chip independence, not deepen Nvidia dependency

• Domestic Alternatives: Chinese chipmakers are being pushed to fill the gap

• The Irony: US loosens restrictions just as China decides it might not want the chips anyway

Meanwhile: ByteDance and Alibaba asked Nvidia about placing large H200 orders after Trump's approval, suggesting not everyone in China got Beijing's memo about self-sufficiency.

Act III: The Smuggling Continues Regardless

The Information reports DeepSeek is developing its new AI model using several thousand Nvidia Blackwell chips smuggled into China via third-party countries:

• Thousands of Blackwell Chips: The latest architecture, theoretically unavailable to China

• Third-Country Pipeline: Chips route through intermediaries to evade controls

• Nvidia's Position: The company says it hasn't seen "substantiation or received tips" about smuggling via data centers

• Location Verification Tech: Nvidia has privately demonstrated unreleased technology that could show which country its chips are operating in

The Enforcement Reality: US DOJ detained two men for allegedly trying to smuggle $160M+ of Nvidia H100 and H200 chips to China; a third pleaded guilty. The arrests prove smuggling networks exist. DeepSeek's alleged access to Blackwell proves it's effective.

Translation: The chip export policy is now simultaneously too loose (Trump's 25% deal), too strict (driving smuggling), and possibly irrelevant (China might reject legally sourced chips while obtaining illegally sourced ones). Everyone's unhappy, which usually means nobody's winning.

🤖 Agentic AI Goes Mainstream: The Protocol Wars End Before They Start

Anthropic, OpenAI, Google, Microsoft, Block, AWS, and others launched the Agentic AI Foundation to build open-source agent standards under the Linux Foundation:

• Unprecedented Coalition: Direct competitors collaborating on shared infrastructure

• Linux Foundation Home: The same organization managing Linux, Kubernetes, and other critical open-source projects

• Open Governance: Development now managed by the broader community

Anthropic's MCP Donation:

• 10,000+ Active Servers: The Model Context Protocol has achieved critical mass

• From Side Project to Standard: The Verge profiled how two Anthropic employees' passion project became the default agent infrastructure

• Permanent Open Source: MCP development moves from company control to community governance

Why It Matters: When OpenAI and Anthropic agree on something, pay attention. When Google and Microsoft also sign on, you're watching infrastructure commoditize in real time. The agent protocol wars ended before most people knew they'd started.

Translation: MCP is becoming the USB standard for AI agents. Just as USB enabled any device to connect to any computer, MCP enables any AI agent to connect to any service. The plumbing layer of agentic AI just became a public utility.

🤖 Product Launch Quick Hits: The Week in AI Releases

🚗 Rivian's Silicon Ambitions: Announced it's building its own 5nm AI chip called the Rivian Autonomy Processor, plus a "Large Driving Model" for fully autonomous driving. The EV maker joins Tesla, Apple, and Google in the "we need our own chips" club.

⚡ Boom's Pivot to Power: Announced Superpower, a 42MW natural gas turbine using supersonic jet engine tech to power AI data centers. When your supersonic plane business needs diversification, becoming a utility company is certainly a choice.

🔌 Google's MCP Integration: Released fully managed, remote MCP servers for Maps, BigQuery, Compute Engine, and Kubernetes Engine. Google joining MCP removes any doubt about which protocol won.

🎨 Cursor's Designer Mode: Launched Visual Editor, a vibe-coding product for designers integrating professional design controls with natural language editing. The designer-developer boundary continues dissolving.

💬 Claude Code Goes Slack: Debuted a Slack integration for Claude Code in beta, letting developers relay coding tasks from where they already spend their workday.

🎬 Runway's World Model: Launched GWM-1, its first world model using frame-by-frame prediction to simulate physics, plus updates to Gen 4.5 adding native audio capabilities.

💻 Mistral's Coding Play: Released Devstral 2 with 123B parameters (requiring at least four H100 GPUs) and Devstral Small at 24B parameters for local deployment. Open-source coding models keep getting bigger and smaller simultaneously.

🔄 Personnel Quick Hits: Musical Chairs, Tech Edition

💼 Slack's CEO Defects to OpenAI: Denise Dresser, Slack CEO since November 2023, is leaving to become OpenAI's Chief Revenue Officer. Salesforce loses its enterprise messaging chief to the company building agents that might replace enterprise messaging.

🤝 Nick Clegg's VC Pivot: The former Meta policy chief joins London-based Hiro Capital as a general partner, with Yann LeCun joining as an adviser. Sources say Hiro is raising a €500M+ fund.

📦 Coupang's Data Breach Fallout: CEO Park Dae-jun resigned after South Korea's largest-ever data breach compromised 30M+ people's personal information. Police raided headquarters.

💕 Hinge Founder's AI Dating Play: Justin McLeod steps down to launch Overtone, an "AI-driven" dating service backed by Match Group. CMO Jackie Jantos promoted to CEO.

📚 Wikimedia's Diplomatic Hire: The foundation named Bernadette Meehan, US Ambassador to Chile from 2022 to 2025, as its new CEO.

🌟 Editor's Note

At Startup Intros, our mission is to bring the latest founder-investor news straight to your inbox, keeping you ahead in the fast-paced world of Silicon Valley.

Introducing the first AI-native CRM

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

• Prospect and route leads with research agents

• Get real-time insights during customer calls

• Build powerful automations for your complex workflows

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉Try Attio Pro for free

💭 Parting Thoughts

The AI infrastructure race revealed its true nature this week: unlimited ambition colliding with limited physics and chaotic policy.

Microsoft and Amazon are committing $60B to data centers. Blue Origin and SpaceX are developing orbital computing to address Earth's power constraints. Meanwhile, 44GW of new capacity doesn't materialize because billionaires want it to; the AI bubble's pin might be a grid that simply can't deliver.

Trump's chip deal upsets everyone: China might reject the chips they're now allowed to buy, while smuggling networks continue to operate. The only clear winner? The Agentic AI Foundation. When every major AI company agrees on standards, the protocol wars end before they begin.

Forward to a friend or hit reply to let me know what you're seeing in your world.

Till next time!

P.S. Raised this week, and we missed you? Want to be featured? Have tips or funding questions? Reply or DM us as we’re here to help.

Startup Intros Trending Thursday: Your trusted source for founder-investor insights, delivered with clarity and focus.