AI Sector Daily Digest: August 10, 2026 — Meta's local model, Unitree's IPO, and the financing test for AI

In brief

1. Meta puts a smaller open-weight model on consumer hardware

• Meta released Muse Glimmer, an open-weight model designed for agentic tasks that can run on a Mac or PC with a single graphics card. The release is a model-availability event, although Reuters did not report a parameter count or a detailed download path. 1
• Mark Zuckerberg used the launch to argue for lower U.S. barriers to open-source AI. Meta also said it plans more models in this direction and will give independent directors power to approve safety criteria for releases. 1
• The practical change is deployment economics: a model aimed at one local GPU can be customized and run without a frontier-lab API, but its real usefulness will depend on the weights, license, and evaluation details Meta makes available. The source did not provide those details. 1

2. Unitree prices a $904 million humanoid-robot IPO

• Chinese robot maker Unitree priced its Shanghai offering at 150.8 yuan per share and plans to raise 6.1 billion yuan, or about $904 million; subscriptions were scheduled to start Monday. If completed, it would be the first humanoid-robot maker listed on mainland China’s exchanges. 2
• Unitree reported nearly 1.7 billion yuan in 2025 revenue, more than four times the prior year, and about 600 million yuan in adjusted net profit. Overseas revenue was more than 40% of sales in each period disclosed in its prospectus. 2
• The business is not yet a broad factory-automation story: Reuters says current demand is concentrated among universities and government-backed projects, while reliability, dexterity, and long-duration autonomous work remain open problems. The IPO therefore gives investors a financing and commercialization signal, not proof of mass deployment. 2

3. Reuters reports North Korean AI tooling for cyberattacks

• South Korean cybersecurity firm Genians said the North Korean group Kimsuky had built large-language-model tools and gathered software that could automate attacks, analyze stolen material, and make phishing campaigns more convincing. 3
• The reported toolkit included Ollama, GPT4All, Msty, retrieval-augmented generation tools, AI-agent frameworks, speech-to-text software, and Cursor. Genians also found finance- and cryptocurrency-themed decoy documents that appeared to be AI-generated. 3
• This is a reported capability, not a confirmed independent finding: Reuters says the claims could not be independently verified. For defenders, the concrete shift to watch is from using AI to write lures toward using models, RAG, and agents inside the attack workflow. 3

4. Data-center lenders add local opposition to the underwriting checklist

• Banks and asset managers still want to finance U.S. data centers, but Reuters reports that they are adding community support, permits, zoning, environmental exposure, insurance, and construction readiness to the project-risk review. 4
• The financing already being arranged is large: JPMorgan and Morgan Stanley helped manage a $12.3 billion bond sale tied to a Meta data-center project in El Paso; Morgan Stanley and KKR Capital Markets arranged a $9.7 billion CyrusOne credit facility; and Bank of America is advising a $16 billion Oracle campus in Michigan. 4
• Data Center Watch counted at least 75 projects worth about $130 billion facing local opposition in the first quarter of 2026. The immediate implication for AI infrastructure is not that financing has stopped; it is that power, permits, and public acceptance can now delay a project before model demand becomes the deciding variable. 4

5. monday.com says AI products are starting to show up in new sales

• monday.com reported $364.6 million in second-quarter revenue, up 22% year over year. It said annual recurring revenue from its AI products doubled from the first quarter and represented 17% of net new ARR in Q2. 5
• The company also reported 2,019 customers with more than $100,000 in ARR, up 37% year over year, and 114 customers above $500,000, up 68%. Those figures cover the broader platform, so they are adoption signals rather than a standalone AI-revenue total. 5
• The distinction matters for investors: AI is contributing to new recurring sales, but the release still reports company-wide revenue rather than a separately audited AI segment. The next useful check is whether the AI contribution keeps rising without weakening the platform's overall margins. 5

The read-through