链捕手ChainCatcher

According to a ChainCatcher report citing Forbes, Andreessen Horowitz and Accel co-led a more than $2 billion Series E funding round for AI coding agent company Cognition. The latest valuation is $48 billion, up significantly from $26 billion four months earlier. More than 30 institutions, including Founders Fund, General Catalyst, Avenir, and Nvidia, participated as follow-on investors.
Cognition’s AI coding agent product Devin generated operating revenue of $492 million in May, which has grown to nearly $900 million. The Information predicts that its year-end annualized revenue will reach between $4 billion and $5 billion.

ChainCatcher report: A research summary released by Sage Road Research for (The AI Trade) says that, so far this year, the Mag 7 have underperformed the Russell 3000 by about 8 percentage points and the MSCI ACWI by nearly 9 percentage points. In July, the CBOE NDX volatility index relative to the VIX reached the highest level since the internet bubble. After entering a correction, the Nasdaq index saw a 4-day rebound of 5%. As of the time of writing, AI companies' share prices are down 20% from their 52-week highs in June.
With AI costs skyrocketing, companies are finding it difficult to realize returns on investment. Uber, Amazon, Meta, and Walmart have all implemented restrictions on employees' use of AI. With model commoditization limiting pricing power, China's open-source models are becoming cheap alternatives to OpenAI and Anthropic. AI capital expenditures have exceeded expectations; by mid-2026, consensus has risen from $527 billion at the end of 2025 to about $800 billion. In 2027, capital expenditures by hyperscale cloud providers are expected to account for 3% of U.S. GDP, more than double the 1.2% peak of late-1990s telecom fiber buildout. Allianz Research calculates that there is a nearly 46% growth gap between AI investment and sales, worse than the 32% gap during the 2001 telecom bubble.

ChainCatcher message, according to monitoring by Bitcoin News, DART stated that it, together with independent white-hat researchers, has recovered more than 50 BTC from wallets affected by the COLDCARD entropy vulnerability and completed the transfer before malicious attackers could steal them. DART is a digital asset recovery organization that works with white-hat researchers to protect funds with vulnerabilities and coordinate their lawful return to their rightful owners.
The white-hat researchers have not requested a bounty and will return the Bitcoin to the owners. The recovered BTC has been transferred to the Crypto Recovery Trust. The trust is a special statutory trust established under Wyoming state law to hold the recovered digital assets, and will distribute them after the rightful owner has been confirmed and verified.

ChainCatcher message. According to SoSoValue data, the spot XRP ETF saw a total net inflow of $12.285 million in a single day. The spot XRP ETF with the largest net inflow is the Bitwise XRP ETF (XRP), with a daily net inflow of $9.3002 million. Its historical total net inflow has reached $609 million.
Next is the Grayscale XRP Trust ETF (GXRP), with a daily net inflow of $2.9848 million, and a historical total net inflow of $145 million. As of the time of writing, the total net asset value of the spot XRP ETFs is $1.506 billion, the XRP net asset ratio is 1.71%, and the historical cumulative net inflow has reached $1.696 billion.

ChainCatcher report: On September 9, 2026, Ara Kharazian, Chief Economist at enterprise spend management platform Ramp, released the 2026 September Ramp AI Index. The index is based on Ramp spending data tracking U.S. enterprises’ AI usage. In August, Anthropic’s share among U.S. enterprises that pay for subscriptions or tokens reached 43.8%, up 0.34 percentage points month over month; OpenAI was 39.8%, up 0.09 percentage points month over month. Overall new AI adoption is still growing, but at a slower pace; technology industries such as information, finance, and professional services are leading.
Head 1%: The median monthly AI spending per enterprise dropped by 9.7%, from $7,976 to $7,205. The effective price per million tokens fell by 41% from the March 2026 peak of $1.15 to $0.68. Lower-cost standard models are driving usage, while frontier models (Opus, Fable, Sol) account for 45% of token share. Open-source models remain limited, accounting for 6.4% of enterprise AI spending and 3.6% across all enterprises.

ChainCatcher message: Ronit Jain, CEO of Liquid Compute, posts an explanation of how compute price hedging works. Anyone who purchases compute, holds GPUs, or lends compute exposure has already taken on exposure to compute price risk. Hedging uses cash-settled contracts that lock in future prices against a reference index, without transferring any GPUs.
Businesses can buy forward contracts to lock next year’s capacity costs into the budget line item, while still being able to purchase from any supplier. Neocloud can sell forward contracts to convert uncontracted hours into a fixed cash flow, enabling lenders to base their lending on it. For the residual value in sale-and-leaseback arrangements, you can roll recent forward contracts once the transaction windows enter the tradable range, or buy forward contracts with put options; the more liquid near-term contracts can be used to pay the option premium.

ChainCatcher message: A report released by The Block Research on September 8 states that between October 2025 and August 2026, during a period when Bitcoin fell by more than 50% and the total crypto market capitalization decreased by more than 2 trillion USD, the total stablecoin supply remained around 290 billion USD, about 90% of which was issued by Tether and Circle.
Over the past 365 days, stablecoin transfer volume exceeded 900 billion USD, more than double compared with 2025. The daily/weekly turnover ratio rose from 0.38x in August 2024 to 0.78x in August 2026. Ethereum holdings are about 147 billion USD (turnover 0.51x/day), Base holds 4.5 billion USD (16.7x/day), Solana is about 13 billion USD (1.08x/day), and Tron is over 90 billion USD (0.25x/day). According to the Bank for International Settlements, stablecoin transaction volume in 2025 was about 3.5 trillion USD, with payments-related activities accounting for 1.1%. Chainalysis data shows that in 2025, illicit addresses received at least 154 billion USD in stablecoins.

ChainCatcher message: a16z crypto research team today released an open-source new version of zkVM Jolt, called Lattice Jolt. This version uses the same architecture as the previous one, but replaces the underlying cryptography from elliptic curves to lattices, achieving post-quantum security, 2–3x faster prover and verifier performance, and a proof size of less than 100 KB.
Lattice Jolt replaces the earlier elliptic-curve-based Dory with a new polynomial commitment scheme, Akita, based on the Module-SIS assumption, targeting full 128-bit security. Akita is led by LayerZero researchers and engineers, and developed in collaboration with Carnegie Mellon University, the University of Southern California, and the a16z crypto team. On the same machine, with CPU alone, Lattice Jolt proving exceeds 2 million RISC-V cycles/sec; with Apple Metal GPU acceleration, it exceeds 10 million cycles/sec, and the prover memory footprint drops to about 200 bytes per cycle.

ChainCatcher report: an economic team at Anthropic built a model to analyze the impact of AI on U.S. employment, growth, and unemployment. The model treats the economy as a bundle of tasks based on U.S. Department of Labor O*NET classifications. AI can keep tasks unchanged, enhance them, automate them, or create new tasks. The U.S. economy is described as having a task-instance value of more than US$3 trillion.
Three scenarios depend on AI capability and adoption speed: a mild scenario similar to the internet era—U.S. GDP increases by 1.6% in 2030 (US$34.1 trillion, 2025 prices), the labor share is 59.4%, and the capital share is 40.6%. In a substantive scenario, AI can perform half of knowledge work, GDP rises by 8.3% (US$36.3 trillion), the wages of knowledge workers are roughly unchanged, and the labor share is 56.1%. In an extreme scenario, AI is more efficient in most knowledge work and nearly fully completes it autonomously; GDP increases by 32.4% (US$44.4 trillion), knowledge worker wages fall by more than 10%, the labor share is 45.2%, and the unemployment rate exceeds typical recession levels.