Zoom's Anthropic Investment Yields About $1 Billion in Profit

Regulatory filings reveal that Zoom Communications' stake in Anthropic, acquired in early 2023, is now worth approximately $1.27 billion — a net profit of about $1 billion. Anthropic is expected to close another funding round as soon as next week, which could push the value of Zoom's stake even higher.

As Anthropic's valuation climbs rapidly, companies that invested early are seeing outsized returns. What stands out is that strategic investments in AI startups are starting to move the valuations of the investors themselves.

NVIDIA Removes Gaming Revenue Category from Financial Reports

NVIDIA has dropped the "Gaming" revenue category from its financial reports, and the change is drawing attention. The company's revenue mix has shifted sharply toward data center and AI, making the move symbolic of how the gaming GPU business — once its mainstay — has lost relative importance.

On Reddit's LocalLLaMA community, commenters are saying the "AI fumble" (a shift in business structure driven by the advance of AI) has gone one step further. It's fair to say the picture of NVIDIA's AI semiconductor business pulling in the bulk of revenue is now even clearer.

US Continues to Weigh Semiconductor Tariffs, None Imminent

Jamieson Greer of the US Trade Representative (USTR) has said the agency continues to consider imposing tariffs on imported semiconductors. Speaking at an event marking Micron Technology's memory chip fab expansion in Virginia, he said there are no plans for tariffs on an "immediate, tomorrow or next week" timeline, and indicated he would keep consulting with the industry.

Tariff policy aimed at promoting domestic chip manufacturing could affect the entire semiconductor supply chain, and industry watchers are following developments closely.

llama.cpp Adds PDL Support for Blackwell GPUs, 5–10% Faster Token Generation

A new feature for NVIDIA Blackwell GPUs, "Programmatic Dependent Launch" (PDL), has landed in llama.cpp. PDL is a technique that improves kernel execution efficiency and works on supported GPUs (Compute Capability 90 or higher, excluding the ADA generation).

Benchmarks on an RTX Pro 4500 Blackwell 32GB show almost no impact on pre-fill, while text generation gained an average of 5–10%. Qwen 3.6 35B-A3B UD-Q5_K_XL recorded a 9.17% speedup.

For now you have to manually enable the `-DGGML_CUDA_PDL=ON` flag at build time, and not every kernel supports it, but for Blackwell users it's a "free lunch" worth trying.

Qwen3.6-35B-A3B Runs 262K Context on an RTX 3070 Ti with 8GB VRAM

A Reddit user has reported a way to run Qwen3.6-35B-A3B (an MoE model) with a 262K context length at around 30 tps (tokens per second) on an RTX 3070 Ti with 8GB VRAM. Because it's an MoE model, only about 3.5B parameters are active at runtime, so even 8GB VRAM can hold the active layers (~3GB), GPU buffers (~2GB), and a 262K KV cache (~2.56GB).

Interestingly, switching to Ubuntu Server delivered roughly a 25% tps improvement over Windows 11 (27→34 tps) — proof that OS overhead has a big impact on inference performance. The setup can also handle 512K context, and in extreme cases reportedly up to 1M.

Virgin Atlantic Overhauls Its Mobile App with OpenAI Codex

In an official OpenAI case study, Virgin Atlantic announced it has completed a full overhaul of its mobile app using Codex. The team developed against an immovable deadline — the year-end holiday travel rush — and achieved nearly 100% unit test coverage with zero P1 defects.

As a real-world example of Codex delivering practical results in an enterprise development project, it's a useful reference for companies considering adopting AI coding tools.

---

**Sources:**

[Zoom's Anthropic Investment Has Netted the Company $1 Billion (Bloomberg)](https://www.bloomberg.com/news/articles/2026-05-22/zoom-s-anthropic-investment-has-netted-the-company-1-billion)

[NVIDIA Removes Gaming Revenue Category (guru3d.com)](https://www.guru3d.com/story/nvidia-removes-gaming-revenue-category-from-financial-reports/)

[US Weighs Chip Tariffs (Bloomberg)](https://www.bloomberg.com/news/articles/2026-05-22/us-weighs-chip-tariffs-to-spur-domestic-growth-trade-chief-says)

[Blackwell and PDL Performance Increase (Reddit LocalLLaMA)](https://www.reddit.com/r/LocalLLaMA/comments/1tkw1su/blackwell_and_pdl_performance_increase/)

[Qwen3.6-35B Q4 262k Context on 8GB 3070 Ti (Reddit LocalLLaMA)](https://www.reddit.com/r/LocalLLaMA/comments/1tkxnuc/qwen3635ba3b_q4_262k_context_on_8gb_3070_ti_30tps/)

[How Virgin Atlantic Ships Faster with Codex (OpenAI)](https://openai.com/index/virgin-atlantic)