South Korea’s two biggest crypto exchange operators reported sharply weaker first-half results on Aug. 14, showing how closely exchange earnings still track trading activity. Dunamu, the parent company of Upbit, posted 408.1 billion won in consolidated operating revenue for the first half of 2026, down 49.1% year over year. Operating profit fell 79.7% to 111.5 billion won, while net profit dropped 74.1% to 108.4 billion won. Bithumb reported 168.8 billion won in revenue, down 48.7%, with operating profit of 14.9 billion won, down 83.4%, and a net loss of 108.7 billion won versus a net profit of 55 billion won a year earlier. The backdrop was a broad contraction in local trading activity. Combined second-quarter volume across South Korea’s five licensed KRW exchanges — Upbit, Bithumb, Coinone, Korbit and Gopax — fell 49.5% year over year to about $146.4 billion, cutting directly into fee income. The report said Dunamu benefited from better cost control, while Bithumb’s loss included digital asset impairment and administrative expenses related to regulatory penalties. It also pointed to a shift in Korean retail money toward AI and semiconductor stocks such as Samsung Electronics and SK Hynix, as well as expectations around a 22% crypto capital gains tax due to begin in January 2027. Both companies are still pushing IPO plans, but their latest numbers put fresh pressure on how public investors may value fee-driven exchange businesses.

Accel’s latest $3.5 billion fundraise, announced on Aug. 11, came just four months after it raised a $5 billion late-stage fund, giving the firm $8.5 billion in fresh capital to deploy into what it calls an AI “supercycle” still in its early innings. The contradiction at the center of the market is becoming clearer: AI tools are helping startups build products and validate business models with smaller teams and less upfront spending, yet the price of buying meaningful ownership in the best AI companies is rising fast. Data cited from Carta show smaller startup teams, lower headcount at later stages, and a funding market that is splitting in two. Lightweight companies can get started with less money, while elite AI startups founded by researchers and executives from places such as Google, DeepMind, and OpenAI are raising unusually large seed rounds at valuations once reserved for growth-stage businesses. Carta and Crunchbase data also point to capital concentrating in a narrow group of leaders, with later-stage rounds gaining share and mega-rounds taking a growing portion of venture dollars. For venture firms, the issue is no longer just getting into a coveted deal. It is having enough capital to keep up as valuations climb and dilution falls.

A new rumor circulating on X has put Safe Superintelligence, the secretive company founded by former OpenAI chief scientist Ilya Sutskever, back in focus. According to posts attributed to user San Zhi Caomei, SSI is working on a small reasoning engine built around test-time training, or TTT, a method that allows a model to keep updating part of itself while solving a problem. The same posts claim the current version is ready, that a next-generation version is being scaled up by 10x, and that an early release could come as soon as August for a limited group of users. None of those claims has been confirmed by SSI or Sutskever, and later chatter suggested a release might not happen this month after all. The rumor gained traction because it appears to line up with several public clues. Sutskever said in a November interview with Dwarkesh Patel that terms such as AGI and pretraining had pushed the field in the wrong direction, arguing that intelligence should keep learning after deployment. Nvidia’s July 27 announcement of a long-term strategic partnership with SSI added to that interest, saying the startup had spent the past two years pursuing a new research path and that its compute would rise by an order of magnitude with access to Vera Rubin systems. In that context, the latest TTT speculation has been read as a possible signal of what SSI has been building behind closed doors.

Engineer Ben Zhang said he spent 30 minutes searching his office for a missing phone before turning to Anthropic’s Claude for help. With Apple’s Find My disabled by MDM, Claude suggested a different route: track the phone through Bluetooth signal strength and write a small utility for the job. According to Zhang, the AI produced the meter in about a minute, and he then walked around the office watching the readings rise until he found the device. Zhang later published the tool, called findphone, on GitHub. The utility is written in Swift, runs on macOS 13 or later, and reads RSSI, or received signal strength indicator, from nearby Bluetooth devices. It can also add a radar-like sound cue through a sound flag and hide Bluetooth addresses with a redact flag for screen recordings. The report notes that RSSI is not a substitute for GPS: accuracy is typically around 2 to 5 meters and can worsen in environments with metal, glass, or water, where reflection and diffraction affect readings. Even so, the episode shows how AI can cut the cost of building one-off software for narrow personal problems.

Jeff Dean, Google’s chief scientist and one of the company’s earliest and most influential technical leaders, has left the company after 27 years to co-found Discovery Loop with longtime collaborators Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. The new company says it wants to automate the experimental loop in machine learning, science, and engineering by using AI models and large-scale compute to generate hypotheses, run evaluations, and iterate at scale. It plans to begin with machine learning research, use that work to improve its own stack, and then expand into areas including hardware design, drug discovery, and clean energy. Discovery Loop is structured as a Public Benefit Corporation. Its seed round was co-led by Radical Ventures and Khosla Ventures, with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet. Financial terms were not disclosed. Google will also serve as a cloud partner and provide at least the first year of compute support. Dean’s departure came as Alphabet shares fell about 5% on the day the news was made public and as Google faces a broader wave of AI talent departures and management changes inside DeepMind and Gemini.

Gavin Baker, founder and chief investment officer of Atreides Management, argued on Invest Like The Best that July’s selloff in AI stocks diverged sharply from industry fundamentals. He said GPU availability, GPU rental pricing, DRAM spot pricing, and token growth were all still accelerating even as many AI names fell 40% to 60% from their highs. Baker said Nvidia is now trading at its lowest forward price-to-earnings multiple of the past decade, which he sees as a sign that public markets are heavily discounting AI earnings. He also said investors misread Meta’s move to rent out compute, overstated the threat from open-source models, and treated widening CDS spreads as a credit alarm when banks may simply have been hedging commitments. At the same time, he identified regulation as the clearest downside risk, discussed long-term memory supply agreements, described Nvidia’s evolving financing model as a form of credit enhancement with revenue sharing, and pointed to SpaceX and orbital computing as underappreciated parts of the broader AI infrastructure buildout.

Safe Superintelligence, the AI startup founded by Ilya Sutskever, plans to release its first model in August, according to Techub News. The company has not launched any product so far, yet it has already raised about $3 billion and reached a valuation of $32 billion. On July 27, SSI said it had entered a long-term partnership with Nvidia, a deal the company said would expand its computing capacity by 10x. CryptoBriefing noted that rising demand for AI training compute could benefit decentralized GPU marketplaces. It added that a fully open release of the model could also support the broader decentralized AI ecosystem. The update links a major centralized AI lab’s expansion in compute access with potential spillover for crypto-adjacent infrastructure tied to distributed computing.

Nvidia said on July 27 that it has entered a long-term strategic partnership with Safe Superintelligence (SSI), the AI company co-founded by former OpenAI chief scientist Ilya Sutskever, and made what it called a “substantial investment” in the startup. The company did not disclose the size of the deal, but Bloomberg reported it was worth about $5 billion, while a source cited by TechCrunch said Nvidia’s investment ran into the billions of dollars. According to PitchBook data cited by TechCrunch, SSI has raised about $7 billion to date and is valued at $32 billion post-money, with backers including Andreessen Horowitz, Alphabet, Lightspeed, GV and Sequoia. The partnership is also designed to expand SSI’s access to computing power. Nvidia said SSI will be able to use its next-generation Vera Rubin computing platform, a move the company said would increase SSI’s compute resources by roughly an order of magnitude, or close to 10x. Sutskever said the company now has research worth scaling, while Nvidia CEO Jensen Huang said he was eager to see what breakthroughs SSI could make on the new platform.