Good morning. It's Wednesday, April 22, and we're covering AI ads moving into chatbots, Apple’s leadership transition, a growing rift over restricted AI use inside U.S. intelligence, and more.
If you enjoy this email, share it with a friend. First time reading? Sign up here to keep up with the future of tech.
SpaceX Partners With Cursor, Secures Option To Buy For $60B
Elon Musk's space and AI empire is making a major play in the coding AI market. SpaceX announced a partnership with Cursor to build a next-generation AI for coding and knowledge work, with an option to either pay the startup $10 billion for the work or acquire it outright for $60 billion later this year. The deal pairs Cursor's developer distribution with SpaceX's Colossus supercomputer — claimed to match the compute of a million Nvidia H100s — and follows xAI recently renting compute to Cursor and hiring away two of its senior engineering leaders.
AI Memory Demand Outpaces Supply As Shortage Persists
Even with billions pouring into new fabs, chipmakers are projected to meet only about 60% of AI memory demand by 2027. A report from Nikkei Asia highlights how hyperscalers and Big Tech firms are consuming nearly all available high-bandwidth memory, leaving supply structurally constrained despite expansion efforts from Samsung Electronics, SK Hynix, and Micron Technology. Analysts estimate output would need to grow roughly 12% annually to close the gap, but forecasts from Counterpoint Research put growth closer to 7.5%, with new capacity not fully online until 2027–2028.
Bond Social Platform Uses AI To Push Users Offline
A new social app is designed not to capture attention, but to actively push users away from their screens. Bond, launched Tuesday by CEO Dino Becirovic, uses AI trained on users’ posted “memories” to generate personalized, real-world activity recommendations instead of maintaining an endless content feed. The platform archives posts privately after 24 hours and learns from user behavior to suggest nearby restaurants, events, or experiences, drawing on signals like past interests and activity patterns.
The NSA Uses Anthropic’s Restricted AI Model Amid Pentagon Dispute
The National Security Agency is reportedly using an AI model its own parent agency has flagged as a potential risk. National Security Agency has deployed Mythos Preview, a restricted system from Anthropic that was withheld from public release due to its offensive cyber capabilities. The model is being used to scan systems for vulnerabilities, despite the Department of Defense labeling Anthropic a “supply-chain risk” after the company denied broad access for surveillance and weapons use.
TFT POLL
The last time you worked alone
We gotta know…
When was the last time you finished a real task (an email, a doc, a deck, code) without touching AI? |
No judgement. Sharing what everyone said tomorrow.
INTERVIEW
Perplexity’s CEO’s Vision of AI Browsing Agents Now Playing Out
"So many people discouraged me from working on the browser. But can you make this decision from offense, not just defense?"9 months ago we interviewed @AravSrinivas CEO @perplexity_ai and it's amazing how much of what he said has already become reality. "Offense is when… pic.twitter.com/G1yNW66n3u
@ForwardFutureView post on XMONETIZATION
AI Chatbots Begin Testing Ads Inside Conversations
AI companies including OpenAI, Google, Microsoft, and Amazon are introducing ads directly into chatbot conversations, marking a shift in both digital advertising and AI monetization. OpenAI, which reportedly expects to burn $25 billion in 2026, is testing ads in ChatGPT to generate revenue from its roughly 900 million weekly users, most of whom use the free tier. Early data shows ads appear in a small share of chats—about 1% on ChatGPT—but are expected to scale quickly, with projections of $2.5 billion in ad revenue this year and up to $100 billion by 2030.
Google’s approach resembles search ads, while ChatGPT places ads later in conversations, aiming for more contextual relevance. Initial results are mixed, with targeting ranging from highly relevant to awkwardly literal, and advertisers still lack robust measurement tools. Despite concerns about trust, users appear largely tolerant so far, continuing conversations at similar rates even when ads appear. → Read the full article here. (Paywall)
SUCCESSION
Apple Names John Ternus CEO as Tim Cook Steps Down
Apple announced that CEO Tim Cook will step down in September 2026 after 15 years, with hardware chief John Ternus set to take over around the next iPhone launch. Cook leaves behind a company that grew into a multitrillion-dollar business but faced criticism for a lack of breakthrough new products beyond iterative updates to the iPhone, Mac, and iPad.
Ternus inherits a highly profitable but maturing company, with challenges including lagging AI innovation, a crowded product lineup, and strained relationships with app developers. The transition comes as Apple leans on external partners like Google for future AI capabilities, highlighting gaps in its in-house efforts. → Read the full article here. (Paywall)
ACCESS
Prisoners Find Workarounds to Access AI Without Internet
In U.S. prisons where internet access is largely banned, inmates are still finding indirect ways to use AI tools like ChatGPT, often relying on friends and family outside to relay prompts and responses. At Maryland’s Jessup Correctional Institution, prisoners have used these methods to draft legal arguments, business plans, and educational materials, sometimes waiting weeks and paying limited funds for message exchanges.
Interest in AI is growing among inmates, with workshops and informal efforts spreading despite strict restrictions officials say are necessary for safety and security. While some see AI as a tool for education and rehabilitation, others warn it could be misused for illicit planning or bypassing monitoring systems. → Read the full article here. (Paywall)
NEWS
What Else is Happening
AI Job Scams Grow More Convincing: Fraudsters use tailored, AI-generated listings and impersonation to trick jobseekers into paying or sharing data, with reports doubling and banks logging sharp increases in victims.
Meta Tracks Employees to Train AI: Internal memos reveal Meta logging mouse movements, keystrokes, and screen snapshots to train AI agents, as part of a broader automation push raising privacy and surveillance concerns.
Google Expands Gemini API: Google is introducing a split approach to AI research tasks, dividing cognitive workloads into distinct analytical and creative personas rather than relying on a single monolithic system.
YouTube Expands Deepfake Detection Access: Platform opens its likeness-protection tool to actors, athletes, and creators, letting them flag and request removal of AI-generated impersonations as deepfakes rapidly spread.
Meta Poaches AI Talent From Murati Startup: Meta hires three more researchers from Mira Murati’s Thinking Machines Lab, including two founders, intensifying Silicon Valley’s battle for elite AI talent.
Volkswagen Adds Voice AI to China Cars: Automaker will roll out in-car AI agents using Tencent, Alibaba, and Baidu tech, enabling voice control and offline functionality as it pushes to regain ground in China’s EV market.
