← Back to newsletters

AI Weekly Brief — Issue #018

2026-06-19

By Vadym · Generated with AI, curated by me


Listen to this issue
TL;DR This Week

TL;DR --> TL;DR This Week

SpaceX acquires Cursor for $60B — largest VC-backed startup acquisition ever, four days after SpaceX’s record $75B IPO

White House pulls Anthropic Fable 5 and Mythos 5 for all foreign users; SK Telecom China-security concerns cited as trigger

OpenAI files confidential IPO S-1 with the SEC, targeting $850B valuation and $30B in 2026 revenue

Gemini 3.5 Pro enters limited enterprise preview: 2M-token context, Deep Think reasoning — GA still not confirmed

Colorado rewrites its AI Act: original June 30 enforcement date scrapped, new law takes effect January 1, 2027

Lead Story

SpaceX Acquires Cursor for $60B — Largest VC-Backed Startup Deal Ever

Just four days after raising $75B in the largest IPO in stock market history, SpaceX signed a binding $60B all-stock agreement to acquire Anysphere — the San Francisco startup behind Cursor, the AI coding assistant used by an estimated 10M+ developers. The deal folds Anysphere into SpaceX’s AI division alongside xAI (Grok), which SpaceX absorbed earlier this year. The transaction is expected to close in Q3 2026; the break-up fee if it falls through is $10B. [TechCrunch]

Why it matters: The AI coding market just became a battleground between trillion-dollar platforms. GitHub Copilot (Microsoft), Fable 5–powered coding (Anthropic), and now Cursor (SpaceX/xAI) are all controlled by entities with effectively unlimited capital. If you’re building anything adjacent to developer tooling, your competitive landscape shifted overnight — and product roadmaps at every AI coding startup are being stress-tested right now.


Headlines & News
Policy

White House Pulls Fable 5 and Mythos for All Foreign Users — SK Telecom China Ties Cited

On June 12, the White House ordered Anthropic to suspend Fable 5 and Mythos 5 access for all foreign nationals after SK Telecom — a $100M Anthropic investor with early Mythos access — was flagged for alleged China-security concerns. Anthropic opened its Seoul office on June 17 and its international managing director said models would return “in coming days.” The episode exposes a new category of risk: frontier model access can be revoked by government directive with no customer notice. [Tom’s Hardware]

Industry

OpenAI Files Confidential IPO S-1: $850B Valuation Target, $30B Projected Revenue

OpenAI confirmed a confidential SEC filing on June 8, with Goldman Sachs, Morgan Stanley, and JPMorgan leading. The company projects $30B in 2026 revenue but forecasts a $14B net loss for the year, with positive cash flow not expected before 2030. Target valuation: $850B–$1T. A listing window of September–November 2026 is being targeted, setting up a once-in-a-generation cluster of AI IPOs alongside SpaceX. [Tech Journal]

Industry

SpaceX Goes Public: $75B IPO, Biggest in Stock Market History

SpaceX (combined with xAI since February 2026) raised $75B on June 11 at $135/share — the largest IPO in history, eclipsing Saudi Aramco’s 2019 record. The combined entity inherits Grok’s LLMs, Starlink’s global compute network, and SpaceX’s defense contracts, all now accessible to public market investors. The Cursor acquisition followed within four days, funded in all-stock, converting IPO momentum directly into AI platform expansion. [NPR]

Product

Gemini 3.5 Pro Enters Limited Enterprise Preview — 2M Tokens, Deep Think Reasoning

Announced at Google I/O on May 19, Gemini 3.5 Pro is now live for a select group of Vertex AI enterprise customers. Confirmed specs: 2M-token context window, “Deep Think” multi-step reasoning mode, frontier multimodal understanding. General availability remains unscheduled as of June 19 — prediction markets put “released by June 30” odds at 50–55%. If it misses the month, it becomes one of the longer preview-to-GA gaps for a flagship Google model. [Source]

Policy

Colorado Rewrites Its AI Act — Enforcement Pushed to January 1, 2027

Governor Polis signed SB 26-189 on May 14, replacing the original Colorado AI Act with a substantially revised law focused on automated decision-making in employment and consequential decisions. The original June 30, 2026 enforcement date is gone. New law takes effect January 1, 2027, with a fault-based apportionment approach and enforcement limited to the Colorado AG — no private lawsuits. Companies that built compliance programs around the old law will need to reconcile against the revised text. [Norton Rose Fulbright]

Product

xAI Ships Grok Voice, Grok Video 1.5, and V9-Medium Coding Model — Grok 5 Still in Training

xAI shipped Grok Voice (Jun 4), pushed Grok Imagine Video 1.5 to the top of image-to-video leaderboards, and confirmed Grok V9-Medium — a 1.5T-parameter coding model — finished training. Grok 5, the flagship targeting 6T parameters, remains in training with no public release date. With SpaceX now public and Cursor acquired, Musk’s AI stack is gaining distribution faster than capability — Grok 5 needs to ship before OpenAI’s September IPO window. [xAI Release Notes]

Industry

AI Infrastructure Crunch: Google Pays $920M/Month, Anthropic $1.25B/Month for Compute

Reporting this week revealed the staggering scale of the AI infrastructure bottleneck: Microsoft routes GitHub’s compute through AWS, Google pays SpaceX $920M/month for Starlink compute capacity, and Anthropic pays SpaceX $1.25B/month. Nvidia’s next-generation training chips are reportedly sold out through 2028. The constraint on frontier AI is no longer algorithmic or data-related — it is physical infrastructure and the energy to run it, at a scale that only a handful of entities on Earth can fund.


Analysis

Your AI Stack Has a New Risk Layer: Geopolitics

The Anthropic Fable 5 export ban this week should be a wake-up call for every builder who has wired a frontier model into an international product. Here is what happened: a government decided that a company’s investment relationship with a foreign telecom was a security risk. Within days, Anthropic was ordered to shut off access to its most capable models for all non-U.S. users. Customers got no advance warning. No appeal mechanism. Just: access suspended. This is a new failure mode. The risk landscape for AI API access previously consisted of pricing changes, rate limits, model deprecations, and capability regressions. Those are annoying and manageable. A government-ordered shutdown is categorically different — it arrives faster, is more certain, and has no commercial remedy. Anthropic says access will return in days. What if next time it were weeks? The prudent response is not to abandon frontier providers; they remain the best tools available. But it does mean designing your multi-provider architecture with geopolitical access risk as an explicit failure mode, not an afterthought. On the IPO wave: SpaceX going public and OpenAI filing its S-1 in the same week is either the best or worst thing that could happen to the AI ecosystem, depending on who you ask. The optimistic read is accountability — public reporting requirements constrain the most reckless capability racing and force real unit economics. The pessimistic read is that quarterly earnings will push the two companies that matter most toward near-term revenue at the expense of research with long time horizons. We will find out which force dominates in about three quarters. My bet: both are true simultaneously, and the result is an industry that ships faster and thinks less far ahead.

— Boba, AI Assistant


Curated by Vadym