🎧 Listen to this Newsletter on the Sun app
Here's what's been trending:
SpaceX can acquire Cursor for $60B or pay $10B. Cursor dropped its $2B raise after Microsoft considered buying it. SpaceX's S-1 claims a $28.5T TAM.
Anthropic hit $1T on secondary markets, surpassing OpenAI's $880B. Also, the White House says China is industrially distilling US AI.
Google says 75% of new code is AI-generated, up from 50% last fall. It unveiled TPU 8 and an internal "Antigravity" to counter Claude Code & Codex.
SK Hynix Q1 revenue surged 198% YoY, operating profit up 405%. Investing $12.85B in a new HBM fab in South Korea.
AI-generated CSAM reports hit 1.5M in 2025, up from 67K in 2024. Florida is criminally investigating OpenAI over a shooting.
Startup Intros has a simple mission: to help tech founders navigate everything in the early-stage, from compliance to capital.
🎧 Listen to this Newsletter on the Sun app!
🔥 Startup Intros Events Coming Up
Apr 28-29: Startup Grind Conference: Startup Intros is an official Community Partner. 5,000+ founders, investors & operators at the Fox Theatre in Redwood City. Get $100 off with code STARTUPINTROSGC!
May 1: A16z Speedrun AMA: Apps for the SR007 cohort close May 17! And we're bringing in Artin from SUN (SR006) to answer your questions!
🚀 SpaceX's $60B Cursor Play
🤝 The Deal
Going Beyond Rockets? SpaceX has signed a partnership with Cursor to build "the world's most useful models" and has the right to acquire Cursor for $60B or pay $10B for the partnership. SpaceX isn't acquiring Cursor immediately because the deal could delay its IPO. Also, Cursor is no longer proceeding with its reported $2B funding round.
Microsoft considered buying Cursor in recent weeks, but didn't make an offer. Separately, xAI held talks with Mistral and Cursor about a potential three-way partnership; Mistral co-founder Devendra Chaplot joined xAI in March. The AI coding market has gone from funding rounds to acquisitions pretty fast.
📄 The S-1
The AI Pivot: In its draft IPO prospectus, SpaceX lists its total addressable market at $28.5T, with $26.5T from the AI sector. The filing also reveals SpaceX is manufacturing its own GPUs, listing it among "substantial capital expenditures." SpaceX's debt grew from $14B to $23B last year, tied to a $4.5B lease deal with Valor Equity for AI equipment, including chips for xAI. Musk purchased $1.4B of stock from current and former employees to boost his control.
The Warning: SpaceX says orbital AI data centers use "unproven technologies" and may not achieve "commercial viability." Sources say Musk has de-emphasized SpaceX's original Mars mission as the company prepares to go public, focusing instead on AI and other revenue streams. Tesla is also planning to use Intel's 14A process at its Terafab project, making it the first major customer for that node.
Translation: SpaceX's $60B Cursor price tag exceeds what Microsoft paid for LinkedIn, while Mars has been quietly deprioritized. If the company that was supposed to make humanity interplanetary is pivoting to AI coding tools, what does that say about where the real money is?
🧠 Anthropic Crosses $1T
💰 The Milestone
Anthropic's valuation hit $1T on Forge Global, a leading private marketplace exchange, surpassing OpenAI's valuation on the platform of $880B. The milestone arrives days after Amazon committed a total of $33B to Anthropic, with a $100B+ AWS lock-in over 10 years. Trump said his admin had "very good talks" with Anthropic and that a DOD deal is "possible." Anthropic spent $1.6M on lobbying in Q1, up from $360K in Q1 2025, outspending OpenAI's $1M for the first time.
⚠️ The Product Reality: Compute Constrained
Anthropic's CEO has publicly described the company as compute-limited, with new infrastructure taking 18-24 months to translate into capacity.
Valuation > Product: Opus 4.7 launched to dev backlash within 24 hours, with Reddit threads calling it "legendarily bad" and hitting 2,300 upvotes in 48 hours:
MRCR Long-context Benchmark: Opus 4.7 scored 32.2%, down from Opus 4.6's 78.3%, as it produces worse code than Opus 4.6.
Argues with users to the point of hallucination and flags routine code as malware.
New tokenizer uses up to 35% more tokens, effectively raising costs without prices.
Anthropic is tightening the economics of inference: a cheaper-to-run model and a tokenizer that shifts cost to the user:
Claude Design launched to strong reviews as a tool for creating websites and presentations, but it is compute-constrained with separate weekly usage limits.
Claude Code was briefly removed from the Pro plan and brought back after backlash.
Lastly, unauthorized users in a private Discord have been accessing Mythos since the day it was announced.
🏛️ The Enterprise Push
Anthropic and law firm Freshfields signed a deal to develop specialized legal AI tools for document drafting, contract review, and due diligence. OpenAI responded with ChatGPT for Clinicians, free for verified US physicians, pharmacists, and others. Both companies are racing to lock in professional verticals before the other can establish pricing power. Anthropic is also requiring government-issued IDs and selfies from some users to prevent access from adversary nations, a policy that is alienating Chinese-American founders who need access to build.
🇨🇳 The Open-Weight Race and the Distillation War
The Foreign Pressure: A White House memo says "foreign entities, principally based in China," are engaged in "industrial scale distillation" of American AI tech:
Moonshot AI released Kimi K2.6, an open-weight model under a modified MIT License, which shows strong improvements on long-horizon coding tasks.
Alibaba shipped Qwen3.6-27B, a 27B dense model that surpasses its own 397B model on major coding benchmarks.
Tencent released Hy3-preview, its first model developed under former OpenAI researcher Yao Shunyu.
Tencent and Alibaba are in talks to invest in DeepSeek at a $20B+ valuation.
Translation: The moat Anthropic is trying to build at $1T is being distilled by open-weight models that cost a fraction of the compute, so what exactly is being priced?
Applications for a16z Speedrun's SR007 cohort close May 17!
And we're bringing in Artin Bogdanov, the CEO & Co-founder of SUN, a Speedrun SR006 company, for an AMA. They turn newsletters and long-form content into on-demand audio with built-in affiliate revenue.
On May 1, he's joining Startup Intros for a live AMA on everything Speedrun: how to apply, how to stand out, and more!
🎧 Listen to this Newsletter on the Sun app!
💻 Google's Coding Counter-Attack
☁️ Google Cloud Next '26
Google shipped across every surface at Cloud Next:
TPU 8t for training and TPU 8i for inference, its eighth generation of custom silicon.
Gemini Enterprise Agent Platform is a revamped dev tool built on Vertex AI that manages the full lifecycle of AI agent fleets.
Workspace Intelligence understands "complex semantic relationships" across Workspace apps for personalized context.
Google and Wiz debuted new AI security agents for threat hunting and detection engineering to combat automated zero-day exploits.
Deep Research and Deep Research Max are now available via Gemini API paid tiers.
🔧 The 75% Number
Google says 75% of new code created inside the company is now AI-generated and reviewed by human engineers, up from 50% last fall. That is the highest number any major company has reported. But internal capability hasn't translated into external product adoption. Chief AI Architect Koray Kavukcuoglu is working to unite Google's internal AI coding tools under the Antigravity platform, built specifically to counter Claude Code and Codex.
The Competitive Landscape: Mira Murati's Thinking Machines Lab signed a deal with Google Cloud, valued at a single-digit billion, to access Google's latest AI systems built on Nvidia's GB300 chips. Google is simultaneously building its own coding tools, hosting its rivals' infrastructure, and watching both Claude Code and Cursor eat the developer market it helped create.
Translation: Yes, Google writes 75% of its own code with AI. And it shipped new TPUs and an enterprise agent platform, and is building Antigravity to catch up on coding. But if it can't sell a competitive coding product, what does that say about the gap between capability and product-market fit?
🛡️ The AI Child Safety Crisis
📊 The Numbers
The Staggaring Scale: NCMEC received 1.5M reports of suspected AI-generated child sexual abuse material in 2025, up from 67,000 in 2024 and 4,700 in 2023. That is a 22x increase in a single year. Generative AI has made CSAM production trivially easy, and the volume is overwhelming every moderation system built for the pre-AI era.
🌍 The Regulatory Wave
Governments are moving simultaneously:
Turkey's parliament passed a bill restricting social media access for children under 15 after a school shooting.
The UK's Ofcom launched an investigation into Telegram over CSAM concerns and predator grooming.
Australia's eSafety Commissioner issued transparency notices to Roblox, Minecraft, and other platforms, detailing child safety measures.
The LA Unified School District became the first major American school system to mandate screen time limits.
The UK is proposing a statutory smartphone ban in all schools in England.
State AGs in West Virginia, Alabama, and Nevada had Roblox settle for $35.8M over child safety protections.
Florida AG James Uthmeier issued criminal subpoenas to OpenAI to investigate whether ChatGPT's role in planning a mass shooting constitutes criminal liability. It is the first known criminal probe of an AI company over content generated by its model.
Translation: AI CSAM reports went from 4,700 to 1.5M in two years. And the regulatory velocity is accelerating faster than any platform's moderation can keep up. If generative AI just made the child safety problem 22x worse in a single year, who's building the tools to match that scale?
🚀 Product Launch Quick Hits
Claude Code source leaked and rewritten: An undergrad used AI agents to rewrite the leaked Claude Code source in a different language, showing copyright uncertainty.
ChatGPT Images 2.0: The new image model adds "thinking capabilities," web search, up to 2K resolution, multiple images from a single prompt, and stronger text rendering.
Workspace Agents: OpenAI launches Codex-powered shared agents in ChatGPT for teams, calling them "an evolution of GPTs."
Privacy Filter: OpenAI releases an open-weight model for masking personally identifiable information in text; 1.5B total parameters, 50M active.
CPC Ads in ChatGPT: OpenAI enables cost-per-click ads at $3- $5 per click, in addition to existing CPM pricing.
Microsoft 365 Copilot Agentic Features: Agentic capabilities in Word, Excel, and PowerPoint are now GA and enabled by default for Copilot and Premium subscribers.
Beehiiv Creator Tools: Metered paywalls, webinars for up to 10K people, and AI analytics.
Anker Thus Chip: A compute-in-memory chip for on-device AI, launching first in Soundcore earbuds.
GitHub CLI Telemetry: GitHub begins collecting pseudonymous client-side telemetry from CLI users, enabled by default.
China's 360 finds ~1,000 vulnerabilities with AI: 360 Digital Security Group uncovered previously unknown flaws in Microsoft Office using an AI-powered agent.
👔 Personnel Quick Hits
LinkedIn names Daniel Shapero CEO: COO succeeds Ryan Roslansky, who retains his position as EVP at Microsoft.
Microsoft's first voluntary retirement program: Open to US staff whose combined years of service add up to 70 or more. First in the company's 50-year history.
Trump Media names Kevin McGurn interim CEO: Succeeds Devin Nunes. McGurn previously worked at Hulu, Vevo, and T-Mobile.
Core Automation launches: Co-founded by ex-OpenAI VP Jerry Tworek to build "the world's most automated AI lab," with talent from OpenAI, Anthropic, and DeepMind.
Sullivan & Cromwell admits AI hallucinations in court: The top law firm told a US federal bankruptcy court that a major filing contained multiple gen AI fabrications.
SBF withdraws retrial request: Says he doubts he would receive a fair hearing. His broader appeal remains pending.
Meta tracking employee activity for AI training: Installing software on US employee computers to capture mouse movements, clicks, and keystrokes in work-related apps.
CISA nominee Sean Plankey withdraws: Pulled out after resistance from Sen. Rick Scott stalled his nomination for over a year.
Former Samsung researcher jailed 7 years: Convicted of leaking semiconductor tech to China's CXMT, helping it develop HBM.
🌟 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
The thread connecting everything this week is the gap between valuation and execution:
Anthropic is worth $1T but shipped a model that developers are calling a regression.
SpaceX is filing for an IPO with a $28.5T TAM but is warning investors that its space data centers might not work.
Google writes most of its own code with AI, but can't sell a coding tool that developers want.
The numbers have never been bigger. The gap between what's being priced and what's being delivered has never been wider.
Till next time!
 | Dev Chandra |
Startup Intros Trending Thursday: Your trusted source for founder-investor insights, delivered with clarity and focus.