Weekly Download #42

In partnership with

Hi !

This week in Silicon Valley, startups & tech:

• OpenAI is Going Superapp with an Ad Platform: ChatGPT is merging shopping, research, and coding into a single product as ads roll out.

• Three Billionaires are Betting $200B+ on Physical AI: Bezos launched a $100B infrastructure fund while Musk unveiled a $20B Terafab factory.

• OpenAI faces Legal Pressure: Microsoft is disputing a $50B Amazon cloud deal that may breach their exclusive partnership.

• EA's $55B is the Largest Gaming Deal Ever Attempted: Saudi Arabia's PIF is leading with Silver Lake on a $20B all-cash debt package.

• Kalshi Hit a $22B Valuation with $1B in Funding: The prediction market's valuation jumped 11x in nine months as its ARR reached $1.5B.

Startup Intros has a simple mission: to help early-stage founders stay informed and navigate fundraising from compliance to capital.

🔥 Startup Intros Events Coming Up

• Mar 30: AI Startup Pitch Night: Founders pitch in front of an incredible group of investors. If you're building in AI, you should apply to pitch here!

• Apr 3: Caffeine & Capital: Our morning coffee meetup is back at Corgi Cafe in FiDi! Grab coffee, meet founders and investors & register on Luma!

🔵 OpenAI's $300B Gambit

🚀 The Superapp Sprint

OpenAI is merging ChatGPT, Codex, and Atlas Browser into a single superapp that handles shopping, tasks, search, and more. ChatGPT now has 900M weekly active users, more than double the 400M from a year ago. The shopping integration was initially planned to support direct in-app purchases, but users will now be routed to third-party apps for payment.

Autonomous Research: OpenAI is building toward an autonomous AI research intern by September 2026 and a fully automated multi-agent research system by 2028. This goal is to integrate work on reasoning models, agents, and interpretability into a single system capable of tackling specific research problems without supervision.

Making Moves: The company is also nearly doubling its workforce from 4,500 to 8,000 by the end of 2026, adding 12 new hires per day. And it acquired Astral, the company behind the Ruff linter and uv package manager, to fold Python developer tooling into Codex.

📢 The Ad Machine

Dave Dugan, former Meta VP of Global Clients and Agencies, joined OpenAI as VP of Global Ad Solutions, reporting to COO Brad Lightcap. Dugan previously managed relationships with the world's biggest ad spenders at Meta and is now building OpenAI's advertising business from scratch.

The Timing: Ads began rolling out to all free and Go-tier ChatGPT users in the US on March 21, using Criteo's ad technology. Higher-tier subscriptions will remain ad-free. With 900M weekly active users, the ad revenue potential is enormous; for context, Meta generates roughly $160 per US user annually from ads.

⚖️ The Legal Walls Closing In

Microsoft is considering legal action over a $50B Amazon cloud deal that may breach the exclusive partnership agreement that came with Microsoft's 27% equity stake in OpenAI's for-profit entity. The dispute centers on OpenAI's Frontier AI product line being hosted on AWS, which Microsoft argues violates its cloud exclusivity terms.

Separately, Elon Musk's lawsuit heads to a jury trial on April 27 in Oakland, seeking $79B to $134B in damages, alleging that OpenAI betrayed its nonprofit mission. On March 13, the Judge called the damages claim "unconvincing" and based on "numbers out of the air." Musk responded on March 17 by pledging to donate any winnings to charity. Microsoft is also named as a defendant for allegedly aiding and abetting OpenAI's breach of fiduciary duty.

Translation: OpenAI is building a superapp with 900M users, launching an ad business to monetize them, and doubling headcount, all while its biggest cloud partner and its original co-founder are threatening to sue it into the ground. OpenAI is betting it can outrun its legal problems with growth. Can it?

🏗️ The $200B Infrastructure Race

Three billionaires committed a combined $200B+ to physical AI infrastructure, each betting that the bottleneck isn't models; it's watts, silicon, and square footage.

🏭 Musk's Terafab

Elon Musk formally launched the Terafab project, a $20-25B joint venture between Tesla, SpaceX, and xAI, announced at the Seaholm Power Plant in Austin. The concept is a factory-of-factories: a single facility that consolidates chip design, lithography, fabrication, packaging, and testing to deliver one terawatt of annual AI compute capacity.

At full capacity, Terafab would produce approximately 70% of the current global TSMC output. Musk is betting that vertically integrating the entire semiconductor pipeline will give his companies independence from TSMC's capacity constraints, the same constraints that have throttled the AI buildout since 2024.

🚀 Bezos Goes All In

Jeff Bezos is raising a $100B fund to acquire and modernize industrial companies using AI. The fund, tied to Project Prometheus (which already has $6.2B in existing funding), targets semiconductors, defense, and aerospace sectors using digital twin technology to create AI simulations of physical systems. It's the single largest individual commitment to AI infrastructure ever announced.

The Connectivity Layer: Blue Origin filed with the FCC for 51,600 satellite licenses under Project Sunrise, which plans orbital data centers on sun-synchronous orbits between 500 and 1,800 km in altitude. The pitch is to reduce terrestrial data center energy and water consumption by moving compute to space, connected by a TeraWave inter-satellite communications network.

🌐 SoftBank & Google Power Up

SoftBank committed to a $33B, 10GW data center campus in Pike County, Ohio, the largest single-site data center project ever proposed. The campus will sit on a former uranium enrichment site, powered by 9.2GW of new natural gas generation with $4.2B in transmission line investment. Construction begins this year with 35,000 workers at peak. Masayoshi Son declared his goal of making SoftBank the largest AI infrastructure company on Earth, as part of a $500B project.

The Power Challenger: Google signed 1 GW of demand-response deals with Entergy Arkansas, Minnesota Power, and DTE Energy, allowing utilities to ask Google to shift or reduce noncritical ML workloads during peak demand. Rather than building more power plants, Google wants the grid to borrow capacity from the companies consuming the most of it.

Translation: Bezos is buying factories to transform them with AI. Musk is building factories to make AI chips. Son is building a campus the size of a small city to run them. These stories should reveal what that tells you about where the bottleneck is.

Global HR shouldn't require five tools per country

Your company going global shouldn’t mean endless headaches. Deel’s free guide shows you how to unify payroll, onboarding, and compliance across every country you operate in. No more juggling separate systems for the US, Europe, and APAC. No more Slack messages filling gaps. Just one consolidated approach that scales.

Get the free guide today

🧠 The Model Race Breaks Open

Three Chinese AI labs made their case this week that the frontier isn't a two-horse race.

🔄 The Self-Evolving Model

MiniMax released M2.7, a model that autonomously handled 30-50% of its own reinforcement learning dev workflow. The system ran an iterative loop for 100+ rounds: analyze failure trajectories, plan changes, modify scaffold code, run evaluations, compare results, and decide what to do with the changes. The model performed its own log-reading, debugging, and metric analysis entirely without human supervision.

The Result: M2.7 scored 56.22% on SWE-Pro (matching GPT-5.3-Codex) and achieved the highest Elo score (1495) among open-source-accessible models on GDPval-AA for document processing. MiniMax is a Shanghai-based lab founded by former SenseTime researchers that IPO'd on the Hong Kong Stock Exchange in January at a $2.5B+ valuation, making co-founder Yan Junjie a billionaire at 36.

📱 The $348 Contender

Smartphone manufacturer, Xiaomi unveiled MiMo-V2-Pro, a 1T-parameter model (codenamed Hunter Alpha) that benchmarks close to GPT-5.2 and Claude Opus 4.6 on agentic tasks:

• ClawEval: MiMo-V2-Pro scored 61.5 (approaching Opus 4.6's 66.3; GPT-5.2 scored 50.0)

• Terminal-Bench 2.0: Hit 86.7. The model landed at #10 on the Global Intelligence Index with a score of 49 (same tier as GPT-5.2 Codex).

The Cost Difference: Running the Artificial Analysis benchmark index costs $348 with MiMo-V2-Pro vs. $2,304 for GPT-5.2 and $2,486 for Opus 4.6 (one-seventh the cost!). The architecture uses a 7:1 hybrid ratio, with only 42B parameters active per forward pass out of 1T total parameters, plus a 1M-token context window.

🔍 The Hidden Foundation

Moonshot AI revealed that Cursor's Composer 2 is built on top of Kimi K2.5, Moonshot's open-accessible model. The discovery came when developers noticed the model ID "kimi-k2p5-rl-0317-s515-fast" in Composer 2's API responses. Moonshot's head of pretraining publicly questioned whether Cursor had complied with Kimi K2.5's modified MIT license, which requires commercial acknowledgment.

The Confirmation: Cursor confirmed it used Kimi K2.5 as a base model through a Fireworks AI partnership, adding that 75% of Composer 2's final compute came from Cursor's own continued pre-training and 4x-scale reinforcement learning. The remaining 25% came from the Kimi base. Cursor acknowledged that omitting the mention was "a mistake" and committed to disclosing sources going forward.

World Models: Packy McCormick and General Intuition CEO Pim de Witte published a 19,000-word essay on the history and potential of world models, AI systems that build internal representations of environments to plan for future events. Multiple labs have collectively raised over $1B in world models research.

Translation: A Chinese lab built a model that trains itself. A smartphone maker shipped a 1T-parameter model that costs 7x less than GPT-5.2. One of the most popular AI coding tools in the US is built on a Chinese foundation. When frontier performance becomes available at a fraction of the cost from unexpected players, what happens to the pricing power of the labs that charged a premium for being first?

🎟️ Come to Startup Grind!

Big news! Startup Intros is officially a Community Partner for the Startup Grind Conference this April!

5,000+ founders, investors & operators at the Fox Theatre in Redwood City, April 28-29. This year, we're planning our own Startup Intros happy hour, so stay tuned!

Get $100 off with code STARTUPINTROSGC!

🎟️ Want a FREE ticket? Refer Us To Your Friends!

We just launched a referral program: Every person you refer to Startup Intros counts as one entry toward a free conference pass.

More referrals, better odds. Winner drawn March 31st!

Every tier gets more personal, every tier gets more valuable:

• 2 referrals: The Ultimate Startup Programs & Accelerators List

• 3 referrals: 5-minute video roast of your website or social media

• 5 referrals: 5-minute Loom pitch deck roast with investor-lens feedback

• 10 referrals: 50 investors list hand-picked for your startup

• 15 referrals: 15 minutes with me for fundraising advice or whatever you need

⚡ Startup Quick Hits

• Kalshi: $1B venture round at $22B valuation led by Coatue Management for the prediction market platform; valuation jumped 11x in nine months as ARR hit $1.5B and weekly trading volume reached $2.9B.

• Cloaked: $375M Series B led by General Catalyst & Liberty City Ventures for a privacy platform that creates unique identities for emails, phone numbers, and passwords.

• Kandou AI: $225M at $400M valuation led by SoftBank for the AI chip company that pivoted from low-latency interconnect to full chip design.

• D-Robotics: $120M Series B1 led by DiDi Global and Meituan Longzhu for the embodied AI robotics computing platform spun off from Horizon Robotics.

• Oasis Security: $120M Series B led by Craft Ventures for agentic access management, securing enterprise AI agents with ARR growing 5x YoY.

• Dash0: $110M Series B at $1B valuation led by Balderton Capital for OpenTelemetry-native observability evolving into autonomous AI monitoring.

• Bluesky: $100M Series B led by Bain Capital Crypto for the decentralized social platform; grown from 13M to 43M+ users since October 2024.

• Latent Health: $80M Series A co-led by Spark Capital & Transformation Capital for the clinical AI platform accelerating medication access.

• RoboForce: $52M Seed led by YZi Labs ($10B fund) for physical AI robots for industrial labor; 11,000+ robot orders via LOI with deep Nvidia partnership.

• Arc Boats: $50M Series C led by Eclipse for electric commercial and defense boats with a $160M contract with Curtin Maritime for hybrid-electric tugboats.

💰 Investor Quick Hits

• a16z Top 100 GenAI Consumer Apps: Olivia Moore's latest report finds ChatGPT is 2.7x larger than Gemini, while Claude paid subscribers are growing 200% YoY.

• Air Street Capital: $232M Fund III for Nathan Benaich's London firm, making it Europe's largest solo GP venture fund.

• Partech: €300M European climate and industrial tech fund targeting B2B companies with €10M+ revenue; 40% already deployed.

• Gradient: $220M fifth fund for the firm spun out of Google; manages nearly $1.2B across five funds with 500+ AI founders backed.

💸 M&A & IPO Quick Hits

• EA $55B Take-Private: Saudi Arabia's PIF leads with Silver Lake and Affinity Partners; $210/share all-cash deal with PIF taking 93% equity and $20B in debt financing.

• Sony-TCL $1B Deal: Sony is selling 51% of its home entertainment division (TVs, home audio) to TCL in a JV; ops are expected to begin in April 2027.

• Grab-Foodpanda Taiwan $600M: Grab acquires Foodpanda Taiwan business for $600M all-cash, which generated $1.8B GMV in 2025.

• Elliott/Synopsys Activist Stake: Elliott Investment Management takes a multibillion-dollar position in chip design software giant Synopsys.

🌟 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.

💭 Parting Thoughts

There are two themes happening at once:

• On one side, the physical layer is becoming the bottleneck: the AI stack is moving from models to agents to chips to concrete and kilowatts.

• On the other hand, frontier performance is becoming available from unexpected places at a fraction of the cost.

Build for a world where compute is scarce but models are abundant. That's the real arbitrage.

Till next time!

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