⚡️Aerospace Brief⚡️ K2 Space raised a $500M Series D co-led by Kleiner Perkins and ICONIQ, reaching a $6.8 billion valuation that more than doubles what the company was worth in December. The round pushes K2 past $1 billion in total funding raised. The company has also crossed $1 billion in signed commercial and government contracts. K2 builds massive high-power satellites designed to pack far more power and payload capacity per spacecraft than conventional platforms. Backers in the round include CapitalG, Lightspeed, Altimeter, Spark Capital, ARK Invest, and T. Rowe Price. @K2Space Source: Bloomberg

AI compute is moving beyond a one-size-fits-all architecture. Our latest deep dive explores how ASICs work, why different AI workloads need different chips, and how hyperscalers and startups are reshaping the economics of AI infrastructure. Read the full report 📜↓

The next robotics interface may look less like coding — and more like turning a volume knob. Enigma @enigma_ai emerged from stealth with a $71M seed round led by @IndexVentures and @RibbitCapital. The company is building AI models and interfaces that make robots easier for humans to control, across different hardware systems. Its first public experiment puts more than 100 real AI-powered robots online, letting users control them from a browser.

NVIDIA @nvidia announced a long-term strategic partnership with Safe Superintelligence, the AI lab co-founded by former OpenAI chief scientist Ilya Sutskever. The official release says SSI will get access to NVIDIA’s next-generation Vera Rubin systems, increasing its compute capacity by an order of magnitude. Reuters separately reported, citing a person briefed on the deal, that NVIDIA will make a $5B equity investment in SSI. The companies did not disclose financial terms. SSI is still a highly secretive research lab, focused on building safe superintelligence rather than commercial AI products today.

AI model access is becoming an infrastructure layer of its own. WSJ reported that Stripe @stripe is in advanced talks to acquire OpenRouter , an AI model marketplace, in a deal that could value the company at around $10B. The transaction is not final, and talks could still fall apart. OpenRouter gives developers one API to access and compare hundreds of AI models across providers, with routing, fallback and usage tracking built in. It sits between developers, model providers and AI spend — where routing decisions can affect cost, latency and reliability.

TechCrunch reported that Encord is working with Zander Labs to test a new physical-AI data pipeline: capturing first-person video and brain-wave signals while humans perform manipulation tasks. This is not about “mind-controlled robots.” The goal is to make training data more informative. Brain signals could help mark intent, error, surprise or cognitive load — context that video alone may miss. Encord’s broader physical-AI data stack also includes in-field collection, teleoperation, multi-sensor data, annotation and deployment feedback. Robots need structured examples of hands, tools, objects, motion, timing and failure cases to learn the physical world.

Physical AI is moving from demos into infrastructure job sites. Gritt has launched with $32.4M in pre-seed and Series A funding to bring robotics and AI into large-scale construction, starting with solar farms. The company’s approach is practical: instead of building a fully proprietary robot fleet, Gritt attaches robotic systems to existing jobsite equipment such as skid steers and forklifts. That lets its machines handle repeatable physical work like pick-and-place, material transport and assembly in outdoor environments where dust, terrain and weather make automation much harder than in factories.

CXMT, formally ChangXin Memory Technologies, began trading on Shanghai’s STAR Market today under ticker 688825. Reuters reported that the company raised RMB 57.92B, or about $8.6B, in Asia’s largest IPO this year. Shares opened at RMB 49.50, compared with an IPO price of RMB 8.66, giving the DRAM maker a sharply higher market value in early trading. CXMT is a major DRAM manufacturer, and memory has become a more strategic part of the AI infrastructure stack as demand for DRAM, HBM and high-capacity storage rises. The listing gives public markets another way to price the memory cycle — not only through established global suppliers, but also through companies expanding capacity into an AI-driven demand environment.

Humanoid robotics is moving into deployment infrastructure. Agility Robotics @agilityrobotics opened a 60,000-square-foot Fremont facility to train, test and improve the AI capabilities behind Digit, its humanoid robot for logistics and manufacturing environments. The site complements Agility’s RoboFab manufacturing operation in Oregon. Fremont will focus on software, new skills, customer workflows, field operations and safety validation. Agility says Digit is already active with Schaeffler, GXO, Toyota Motor Manufacturing Canada and Mercado Libre, with more than $300M in conditional multi-year orders for Digit v5. For Physical AI, the infrastructure around the robot may become as important as the robot itself. Source: Agility Robotics / Business Insider

After the World Cup brought Kalshi @Kalshi and Polymarket @Polymarket into a larger sports audience, the next debate is not only about volume. It is about market integrity. The Atlantic described a new risk as “outsider trading”: when traders may try to influence the real-world event or data point that determines a market outcome. One example: French authorities investigated possible tampering with a weather sensor at Paris Charles de Gaulle Airport after an unusual temperature spike affected a Polymarket weather contract. Axios also reported that several Arizona government entities have moved to stop employees from using nonpublic information to profit from prediction markets. That is the next layer of the category.

AI startups are securing compute before the model race is fully visible. Reuters reported that Reflection AI @reflection_ai signed a more than $1B compute-capacity deal with Nebius @nebiusai, including access to Nvidia’s latest chips. Reflection was founded by two former Google DeepMind researchers and is building open AI models as an alternative to closed frontier labs. The Nebius deal follows Reflection’s earlier compute agreement with SpaceX, which TechCrunch reported could be worth up to $6.3B through 2029.

Travis Kalanick's ATOMS just raised $1.7B led by a16z, one of the largest funding rounds in robotics and Physical AI this year. As Travis wrote: "Industrial AI is getting ready to power the next Industrial Revolution." It's exciting to see what this means for robotics, automation, and the future of the physical world. 🚀

Prediction markets had their breakthrough moment at the 2026 World Cup. ⚽ Kalshi: $22.4B World Cup trading volume ⚽ Polymarket: $14.3B trading volume ($10.8B global + $3.5B U.S.) ⚽ New platforms launched, institutions joined the market, and user growth accelerated. Read our latest report on the impact and key takeaways from the 2026 FIFA World Cup.👇

After Spain won the 2026 final, Kalshi @Kalshi and Polymarket @Polymarket showed how prediction markets can turn major sports events into large-scale trading activity. CNBC reported that Kalshi added 3 million new users during the World Cup. The company also said more than $1.2B was traded on its World Cup winner market, making it the largest single market in Kalshi’s history. Earlier in the tournament, CNBC reported that Kalshi reached more than $31B in June notional volume, up more than 70% from May, with sports contracts accounting for roughly 85% of activity. Apptopia reported that Kalshi and Polymarket together held 73.5% of betting-market app downloads in the first half of June. The World Cup put prediction markets in front of a mainstream sports audience.

AI chip competition is moving deeper into inference. WSJ reported that Etched @Etched is in talks to raise capital at a valuation of about $20B, led by existing investor Jane Street. The company is also raising a separate Sequoia-led round at a reported $10B valuation. Neither financing has closed, and terms could still change. The reason investors are watching is clear: Etched is building Sohu, a chip system designed specifically for transformer inference — the part of AI that runs models after training. Etched says it has already raised $800M, built working silicon, and signed more than $1B in customer contracts for its frontier inference clusters.

Sunday Robotics @sundayrobotics released ACT-2, the model powering its home robot Memo. In its preview, Sunday reported a 99.1% ±0.3% success rate across 785 autonomous laundry-folding attempts, covering 9 garment types in unseen home environments, with zero home- or garment-specific adaptation at deployment. For home robots, the hard problem is not whether a robot can complete a chore once on camera. It is whether the same capability holds across different rooms, fabrics, lighting conditions and starting positions. Sunday calls this a “Solve”: reliable performance within a clearly defined scope and adaptation cost. Source: Sunday Robotics / Business Insider

Physical AI needs more than better robots. It needs real-world training data. Business Insider reported that Munich-based robotics startup microagi has raised a $55M seed round, described as the largest seed round ever secured by a German startup. Microagi is not trying to build its own robot hardware. Instead, it trains existing robotics models with footage of real workers performing physical tasks, captured through cameras and sensor-equipped gloves. Its data-collection arm, shift, now operates in 15 countries and pays more than 20,000 people to record real-world tasks. Source: Business Insider

AI startups are moving from fundraising headlines to product tests. Thinking Machines Lab @thinkymachines, founded by former OpenAI CTO Mira Murati, has released Inkling, its first open-weights AI model. The company says Inkling is built as a customizable foundation model. It can be fine-tuned through Tinker, Thinking Machines’ platform for adapting models to specific workflows. The next AI platform battle may not only be about who builds the most powerful model. It may also be about who gives companies more control over how AI is customized, deployed and used. Source: Thinking Machines Lab / WSJ

@Does a robot need to look human, or does it need to fit the job? Business Insider reported that some Silicon Valley investors are pushing back on the humanoid-first thesis, arguing that robot design should start with deployment scenarios — not the human body. The article points to several real-world use cases: warehouse and logistics work, factory tasks, retail shelf scanning, household chores such as clearing tables or loading dishwashers, and industrial robots designed around wheels or specialized bodies. Even @Apptronik offers a useful example. CEO Jeff Cardenas told BI that while he sees the biggest long-term potential in legged humanoids, the company’s wheeled version of Apollo could be deployed sooner because it poses fewer safety challenges. Humanoids may offer broader long-term flexibility, but task-first designs may reach commercial deployment faster. Source: Business Insider

The Wall Street Journal reported that Nscale @nscale secured a $900M revolving credit facility from a consortium of global banks to accelerate data center expansion across Europe, the U.S. and Asia-Pacific. Nscale is a U.K.-based AI infrastructure startup backed by Nvidia, Dell Technologies and Nokia. It provides GPU capacity for AI training and deployment through its data center network. AI compute is beginning to look less like a pure software story, and more like long-lived infrastructure — with large upfront capex, physical sites and bank-backed financing.