Initialized Capital

It's been a delight to watch the efforts of @rahulchhabra07 and the @sabicap team. Feels like a window into the future.

@greptile just got a brand refresh done entirely in-house. we pulled inspiration from moments in the past that once felt futuristic, from the terminal to video games, 80s malls, neon genesis, and of course reptiles, their repeating scales, and their colorful environments. the result is a diverse color palette, intricate illustrations, and modular geometric forms that feel complex yet precise and ethereal. 🧵the process and thinking below:

Full Azura episode on Spotify: open.spotify.com/episode/1IgPbZ…

Full Azura episode on YouTube: youtube.com/watch?v=yonjuO…

@brettdg talks with @jacksondenka, founder of @AzuraTrade, about what it actually took to build an onchain trading platform: cross-chain execution, rolling their own infrastructure, and why app revenue has gone from less than 5% to 60% of all DeFi revenue since he started building. Try it: app.azura.xyz Full episode ⬇️

A startup called Sabi just came out of stealth with a beanie that reads your thoughts. 70,000 to 100,000 miniature EEG sensors woven into the fabric. You put it on like a winter hat and type by imagining the words. Sabi's bet is that if you throw enough sensors at the skull, you don't need to open it. They've already logged 100,000 hours of brain data from 100 volunteers to train a "brain foundation model." First product ships by the end of 2026. A baseball cap version is next. Backed by Vinod Khosla, who also wrote an early check to OpenAI. Khosla: "The biggest and baddest application of BCI is if you can talk to your computer by thinking about it. If you're going to have a billion people use BCI for access to their computers every day, it can't be invasive." The first consumer BCI won't look like a medical device, but instead like something you already own.

you can now control things with your brain. literally. we're building the most wearable BCI on the planet, with @sabicap, backed by @khoslaventures @accel @initialized & @kevinweil. we collected the world’s largest neural dataset and trained the most capable Brain Foundation Model. then we invented a new class of biosensors powered by custom ASICs. type without typing. click without clicking. a cap that lets your brain do the work. we’re sabi.

new brand

Sygaldry @sygaldry_tech Raises $139 Million to Build Quantum Computers for AI. Sygaldry, envisioning a future of affordable, sustainable, personalized, and secure superintelligence, led by Chad Rigetti @ChadRigetti, Idalia Friedson @IdaliaFriedson, and Michael Keiser, has raised $139M in Series A and Seed financing to build quantum-accelerated AI servers. The $105M Series A in March 2026 was led by Breakthrough Energy Ventures @Breakthrough, following a $34M seed round led by Initialized Capital @Initialized. Sygaldry’s servers aim to exponentially speed up critical AI algorithms, reducing the cost and power needed to train and operate increasingly large models.

People ask why we care so much about benchmarks @grepdotai. Because our customers use Grep for serious work like corporate due diligence, Underwriting decisions, and RFP responses for contracts, where the decisions cost hundreds of millions. When the stakes are that high, "pretty good" accuracy is a liability. Here's what #1 actually means. 🧵

This is the viable path for large software outcomes now: figure out how to win the distribution, be AI native, be a compound startup.

Massive congrats to Chad, Idalia, Michael, the entire Sygaldry team—and to our co-investors, old and new! Less than a year ago, I led Sygaldry’s seed round at Initialized Capital, and it's been a thrill to watch them build at terrific speed. Given current trajectories, AI researchers will hit a physics-and-power wall long before they hit a software wall. Sygaldry’s answer? A quantum-classical server that drops into existing AI data center infrastructure and attacks quantum-amenable AI workloads faster and more efficiently than GPUs alone. The architecture combines multiple qubit types in a single fault-tolerant design, purpose-built for the cost, scale, and speed requirements of production AI. The goal, as Chad puts it, is a fundamentally more efficient way of converting megawatts into intelligence. Accelerated research throughput on quantum-friendly workloads and meaningfully improved blended energy efficiency rate across the full deployment. A profound power-up within humanity's most ambitious project. Honored to be onboard with Breakthrough Energy Ventures, Y Combinator, Rock Yard Ventures, RRE Ventures, and others joined Initialized in backing Sygaldry’s vision! LFG.

@ChadRigetti, @IdaliaFriedson, and Michael Keiser are building quantum-accelerated AI servers for the data center, taking a quantum-native approach to one of the hardest problems in AI infrastructure: $5.2 trillion in capex needed by 2030, 125 GW of new power generation to meet global AI demand. They've raised $139M across their seed and Series A. We led the seed, @Breakthrough led the Series A. If you want to work with this extraordinary team on this problem, they're hiring across the board in Ann Arbor and SF: jobs.ashbyhq.com/sygaldry-techn… Full story from @agarfinks at @FortuneMagazine: fortune.com/2026/04/14/exc…

We're officially on the board!

Honestly, this rings true in practice too, I've found I have moved all my research type prompts to Grep.

new robot dropping soon. absolute game changer.

This is what Grep looks like in action when doing ultra deep research...

Grep just achieved SOTA on the three major deep research benchmarks, beating Perplexity, Google, Nvidia, OpenAI, and Anthropic. We're a two-person founding team.

you can now get devin to fix issues flagged by greptile with one click, launch a devin session that includes the issue and relevant context. greptile and devin will loop until greptile gives the PR a 5/5 this has become the default way in which i use greptile

Earthset. This photo from Artemis II would have taken 6 days to send on Apollo's S-band radio. @NASA and @MITLL beamed it back from the Moon in seconds. On a laser. More bandwidth needs higher frequencies. This is Shannon's theorem. It's why 3G → 5G mattered. And that's why lasers beat radio. Laser comms are hard. It's a game of insane precision. The beam starts 10cm wide. By the time it reaches Earth it's 6km across. But from 384,000km away, on a moving spacecraft, that 6km circle is a tiny target. You need to aim to within a thousandth of a degree. The diffraction limit is a physics limit that dictates the minimum feature you can resolve or target as a function of wavelength and distance. Engineering close to the limit is insanely hard — requiring tiny mirrors and active feedback loops that make adjustments faster than the environment can disturb them. Only a handful of organizations have ever built machines that get close. @ASMLcompany prints 8nm features onto silicon with EUV light, and built a ~$500B near-monopoly doing it. (Btw, congrats to the team at @substrate, who just published results printing 12nm features with X-rays). MIT and NASA hit a 6km target from 384,000km away. Different applications, the same physics. NASA has actually had a space internet since 1963 — the Deep Space Network. Three giant dish complexes in the Mojave Desert, outside Madrid, and near Canberra, 120° apart so something is always in view. It brought us pictures from Mars, Jupiter, Saturn, Pluto and much more. Until now it was RF-only. Now it's optical too. At @arenaphysica we're building a foundation model for electromagnetism — starting with RF, architected to generalize to optics as it scales. We exist to help ambitious teams solve hard engineering and applied physics problems like this. And we're hiring. lnkd.in/e_3bXcRV