AI Weekly Malaysia

DeepSeek open-sourced DeepInfer, its production inference framework for DeepSeek-V4.1-Flash, released on day one of PyTorch Conf. It is a heavily modified vLLM fork (crediting SGLang and TRT-LLM) that supports up to 72-hour prompt caching versus the 12 hours OpenAI announced, but the release withholds DeepSeek's ultra-compact KV cache, prefill/decode disaggregation, and load balancing. DeepSeek's own translated figures claim 155T tokens/day per cluster, ~1.5M cache-inclusive input tokens/sec per prefill GPU with a 96% cache hit rate, 14,000 output tokens/sec per decode GPU across ~2,400 GPUs, and >80% margin at $5-8 all-in cost per card-hour against $0.15/$0.60 per million input/output tokens.

Why: The 72-hour prompt caching is the part you can actually plan around: long-running agent sessions and multi-day RAG workflows stop paying repeated prefill costs, which changes how you'd design context retention versus rebuilding state each run. But do not benchmark your stack against DeepSeek's published throughput — the KV cache, PD disaggregation, and load balancing that produce those 14k tok/s decode numbers are explicitly not in the repo, so a vanilla DeepInfer deploy will land well short. Treat the $22m/day and 81-day payback figures as self-reported vendor economics, not a reproducible result.

Spotify launched paid podcast memberships on Tuesday, rolling out in 34 markets including the U.S. Creators choose perks such as bonus audio/video, fewer ads, early access, or combinations across multiple tiers, and keep 90% of membership revenue. Eligibility requires hosting on Spotify for Creators or Megaphone, 100 consumption hours, 100 listeners/viewers, and at least three published episodes; purchases happen on the web, not via in-app purchases.

Why: Only actionable if you build podcast/creator monetization or paid-community tooling: Spotify is adding a paid-tier surface with a 90/10 revenue split and web checkout, so integration or competitive analysis is warranted. For most developers, AI learners, and SaaS founders, this is media-platform news with no code, API, pricing, or Malaysia/SEA detail to act on.

CNBC's Oct. 23 Morning Squawk says a tech-led selloff, led by Meta and chip stocks, gave Nasdaq its worst day since June and pushed the 10-year Treasury yield below 5% with its biggest one-day drop since March 2023. The safe-haven move extended into Japan through surging Japanese government bonds and the yen; the excerpt says Japanese core inflation hit its highest level but cuts off before giving the figure. It also cites investor concern over AI-related bond issuance as part of the rotation.

Why: There is no direct developer-tool, cloud-pricing, or Malaysian builder action in the text. The one decision-relevant detail is that investors are citing AI-related bond issuance as a concern; without specifics on capex, funding terms, or cloud prices, treat this as macro context rather than a reason to change build plans.