Future Technology[Weekly Roundup] SpaceX bought your code editor for $60bn
Plus Google's brain drain, a 2-million-token Gemini, OpenAI's own chip, and a decade-old botnet busted.
Five stories ran the week: a record-breaking acquisition, a quiet raid on Google's best minds, a Gemini that can swallow a whole codebase, OpenAI building its own silicon, and a decade-old botnet finally pulled apart. Here is what mattered, and why.
This week's top stories
1. SpaceX bought Cursor for $60 billion
SpaceX is acquiring Anysphere, the company behind the AI coding tool Cursor, for $60 billion in stock. It is the biggest acquisition of a venture-backed startup on record, and it landed four trading days after SpaceX's own $75 billion IPO, the largest in history. The plan is to fold Cursor into Musk's AI company, xAI.
Why it matters: The real prize is not the product, it is the data, compute and engineers. Every keystroke in Cursor becomes training fuel for Grok, and if your team uses it, your codebase now sits one step from xAI.
2. Google's best AI minds are walking out the door
Noam Shazeer, who co-invented the Transformer architecture every modern chatbot runs on, left Google for OpenAI on June 18, less than two years after Google paid $2.7 billion to bring him back. Days later, AlphaFold lead John Jumper reportedly departed for Anthropic. Two of the most important researchers alive, gone in a week.
Why it matters: The pool of people who can do original, production-grade AI research is tiny, and right now it is draining away from Google toward its rivals. Labs now treat a single hire as a strategic event.
3. Gemini 3.5 Pro can now read 2 million tokens at once
Google's Gemini 3.5 Pro is reaching general availability with a 2-million-token context window, the largest in any production model. That is roughly the entire Harry Potter series in a single prompt. A new Deep Think reasoning mode ships alongside it, gated behind the $250-a-month Ultra tier.
Why it matters: The race has shifted from whose model is smartest to how much it can hold in its head, and how much the best features cost. Frontier AI is becoming something you rent at a premium, not a baseline everyone gets.
4. OpenAI built its own chip to spend less on Nvidia
OpenAI revealed Jalapeno, its first custom inference chip, designed with Broadcom to run its existing models more cheaply. It arrived in a week where SpaceX signed a $6.3 billion compute deal with startup Reflection AI and memory makers posted record demand.
Why it matters: Inference is the bill that never stops when you serve hundreds of millions of users. Building your own silicon is the only real way to loosen Nvidia's grip on pricing, and everyone at the front of the race is now doing it.
5. Police seized 106 servers from a decade-old botnet
Authorities disrupted SocGholish, a malware operation tied to Russian group Evil Corp, seizing 106 servers and cleaning around 15,000 hijacked WordPress sites that had been pushing fake browser-update malware. It has been a front door for ransomware and data theft for years.
Why it matters: If you run a WordPress site, you could be the unwitting middleman in an attack like this. Patch WordPress and its plugins now, and treat any pop-up telling you to update your browser as hostile. Real updates never arrive through a website.
One to watch next week
OpenAI's chief scientist has called GPT-5.6 a meaningful step up from 5.5, with talk of a 1.5-million-token context window. If it lands, the context-length race gets another twist. All times worth checking in your own zone, since these launches tend to drop on US West Coast afternoons.
Forwarded this? Get your own free briefing at futuretechnologyhq.com/newsletter.
That is your week. Which of these still matters in five years, and which is just noise? Hit reply and tell me. I read everything.
Nath, Future Technology
Some links in this newsletter may be affiliate links. We only recommend products we genuinely think are worth your time.