MediaTek commits $5B: the Taiwanese fabless giant enters the AI capex league
The board approves a significant AI-focused budget—first custom AI chip expected in Q4 2026, a second in 2028.
yesterday 12 13
Nscale/Anyscale: Ray becomes the neocloud's new customer lock-in
Nscale acquires the company behind Ray - the open-source distributed-compute framework ML teams actually write against. The prize is placement, not code.
Jul 30, 2026 at 19:38 11 14
Huawei's Thailand AI push: China's soft-power stack finds its next customer
Huawei Cloud makes its "Agentic Infrastructure" available in Thailand and opens the beta of CodeArts Agent in the country. Chinese AI diplomacy (thread `china-ai-soft-power`) moves from discourse to product deployment.
Jul 26, 2026 at 18:40 9 8
Synthetic customers, real budgets: how one startup swapped focus groups for AI personas
A Singapore-based startup is replacing expensive focus-group panels with AI-generated personas that "respond" like segmented buyer archetypes. It's fast, it's cheap - and it's about to trigger a real methodology fight in market research.
Jul 11, 2026 at 11:41 4 7
"Cancelling Cursor": Quality debt takes precedence over feature velocity
Alex Yumashev (Jitbit) explains why he unsubscribed from Cursor: VSCode base 7-8 months behind, 7 bugs reported in a year and never fixed, Agent mode forced with every update, ghost support. An individual signal that perfectly illustrates the editor-orchestrator divide.
17 h ago 11 10
Opus 5 reduces the prompt injection rate to 2% at k=15 - Schneier validates the trajectory
On the IPI benchmark, Opus 5 reduces the attacker success rate of Opus 4.8 by nearly threefold. The best non-Claude model evaluated remains at 16.5%. Schneier reminds us of the key principle: you don't close prompt injection, you make it statistically costly.
Jul 31, 2026 at 22:20 10 11
Refactoring as a token-cost lever: an experiment in Fowler's gen-AI series
In Fowler's exploring-gen-ai series, Giles Edwards-Alexander runs a small experiment: decompose a large function, and watch what happens to the token cost of AI-assisted changes. The lever refactoring gives you is now measurable in dollars.
Jul 30, 2026 at 19:40 16 18
Hugging Face used to strip women and children - the cost of open weights is on the platform now
A report documents how Hugging Face is used to easily create non-consensual intimate images - including of children. The cost of open weights is no longer theoretical.
Jul 28, 2026 at 21:00 10 12
China Telecom deploys the first domestic TPU cluster - Zhonghao Xinying brings silicon to service
China Telecom deploys the first TPU cluster in Hangzhou based on Zhonghao Xinying's Chana chips. The Chinese sovereign compute gains a third silicon vendor with a scale customer.
Jul 28, 2026 at 21:00 17 12
SK Hynix Q2 record: memory is where the AI capex pinch prints first
SK Hynix aims for a record Q2 profit - 57% of the HBM AI-server market according to Tech in Asia - and becomes de facto the 2nd largest shareholder of Kioxia, without voting rights for now. The HBM constraint becomes the clock of the token cost 2027-2028.
Jul 26, 2026 at 18:40 12 8
Code Mode in production: "26 tool calls, one script, $0.02" prices the end of the tool loop
Agent Swarm publishes the cleanest production measurement to date: Code Mode delivers a model-invariant ~99.2% dollar cost reduction versus the classic tool loop. Cloudflare (99.9% token cut) and Anthropic (98.7% on Drive→Salesforce) had promised it; it's now reproduced on real workload.
Jul 23, 2026 at 18:39 6 6
A researcher finds a WordPress RCE for $500,000 with GPT-5.6 for $25
The marginal cost of discovering a critical exploit falls below the threshold of an evening of research - the discovery/reward market ratio explodes.
Jul 20, 2026 at 16:39 19 14