Let me start with a number that made me do a double-take: $60 billion.
That’s what SpaceX just paid to acquire Anysphere, the company behind the AI coding platform Cursor. All-stock deal. Expected to close in the third quarter.
I’ve been watching this space for years, and I’ll be honest—I didn’t see this coming. A rocket company buying a coding assistant? It sounds strange until you realize what’s actually happening.
SpaceX’s AI business, built around xAI, just got a massive injection of talent and technology. Cursor is one of the first enterprise AI products to generate substantial revenue. By bringing it in-house, SpaceX isn’t just buying a tool—they’re buying a beachhead in the enterprise AI market.
But here’s the thing: that’s not even the most interesting story from this week.
The AI That Sees What You’re Thinking
While SpaceX was making headlines, a team of researchers in Turkey was quietly solving a problem that’s been bugging neuroscientists for years.
They built an AI called NeuroSilentia that can decode imagined speech from EEG signals. In plain English? It reads your thoughts. Well, sort of.
The challenge has always been that everyone’s brain signals look different. When you think of the letter “A,” your EEG pattern is unique to you. What works for one person doesn’t work for another.
The Turkish team cracked this by forcing their model to ignore who the signal came from and focus on what letter it represented. The result? Their accuracy jumped from 20% to 51%—and up to 60% with cleaner data.
This isn’t ready for clinical use yet. The lead researcher, Safa Dörterler, was honest about that. But it’s a meaningful step forward. For patients who’ve lost the ability to speak—stroke survivors, people with paralysis—this could eventually give them a voice again.
The Memory Problem Nobody Talks About
Here’s something most people don’t realize: AI models are incredibly wasteful with memory.
Self-driving cars, smartphones, humanoid robots—they all compress images to save space, but that compression means losing track of tiny objects, fine edges, and minor defects. Processing everything in high resolution requires too much power for mobile devices to handle.
Researchers from KAIST, MIT, and Microsoft just solved this with a technique called Upsample Anything.
The beauty of it? It restores fine visual details from compressed, low-resolution data without requiring additional training. It analyzes the boundaries and colors of a single image and calculates the best way to reconstruct lost details. For a standard image, the process takes about 0.4 seconds.
And here’s the kicker: it uses up to 16 times less memory.
Professor Kim Chang-ick, who led the research, put it perfectly: “This technology is an algorithm that can significantly increase the visual precision of artificial intelligence with minimal resources, and we expect it to accelerate the practical application of humanoid robots and on-device AI”.
This isn’t just an academic win. It was recognized at the CVPR conference not just for performance but for computational efficiency and research transparency. That’s rare.
The Protein Puzzle
And then there’s the one that genuinely surprised me.
Researchers at the Chinese Academy of Sciences developed a generative AI model called Void-X that predicts protein-protein interactions at the atomic scale.
Proteins are the workhorses of the human body. They build tissues, transport molecules, regulate cellular communication, defend against infection. Many medicines—cancer therapies, insulin—work by interacting with specific proteins.
The ability to predict and engineer how proteins interact could open new possibilities for treating disease.
What makes Void-X different is its approach. Most AI protein design frameworks start with an overall protein scaffold and then design sequences to optimize binding. Void-X takes a bottom-up approach, generating atomic clusters optimized for tight packing within specific structural regions.
The model contains 172 million parameters and achieved predictive accuracies of 78.3% for intra-chain clusters and 68.2% for inter-chain clusters.
This isn’t just academic. It has real applications in drug discovery and synthetic biology.
The Bigger Picture
Three stories from one week. A rocket company buying an AI coding platform. An AI that reads your thoughts. A model that predicts protein interactions at the atomic scale. And a memory technique that could make AI run on devices we already own.
What ties them together?
AI is becoming infrastructure.
Not just the models we chat with. The systems that run underneath everything. The tools that help us build. The algorithms that help us understand ourselves. The techniques that make it all possible on the devices in our pockets.
The companies winning aren’t the ones with the biggest models or the flashiest demos. They’re the ones building the foundations.
SpaceX buying Cursor isn’t about coding. It’s about control.
KAIST’s Upsample Anything isn’t about images. It’s about making AI work everywhere.
Void-X isn’t about proteins. It’s about understanding the building blocks of life itself.
This is the week AI stopped being a toy and started being the infrastructure we’ll all build on.
Pay attention.
AI, #SpaceX, #Cursor, #Neuroscience, #BrainAI, #KAIST, #ProteinDesign, #DrugDiscovery, #Innovation, #TechNews, #Infrastructure, #FutureOfWork, #AIMemory, #GenerativeAI