← AM·PM Brief

DeepSeek Slashes Prices 75% as Inference Costs Bite, White House Eyes Open-Source Caps to Counter China's AI Lead | AI Daily Brief

July 13, 2026 · Evening brief · 10 news · 6:02

Audio in Mandarin Chinese · English transcript below

⚡ DeepSeek slashes prices but Agent costs spike; White House targets open-source, Nolan mocks AI hype, community fights back with labels, musicians lose gigs to AI, Hassabis chases scientific dreams, medical LLMs self-teach pathology, AI begins self-research, US-China compute race intensifies

Xiao Su and Alex Xiu's AI Daily Brief this week dissected two seemingly unrelated developments that, taken together, reveal the deepening predicament of contemporary AI. DeepSeek's aggressive price cuts, far from resolving the "100x problem" plaguing Agent systems, have instead laid bare the structural fragility of traditional SaaS cost models—what appeared to be democratization may merely be a race to the bottom that exacerbates rather than alleviates the economic unsustainability of inference-heavy architectures. Meanwhile, the White House's deliberations on capping Chinese open-source models, triggered by China's apparent lead in this domain, has cast a geopolitical shadow over the open-weights movement that once promised to be AI's great equalizer. The juxtaposition is telling: as the industry exhausts itself in a war of attrition between plummeting inference costs and escalating policy maneuvering, Demis Hassabis's insistence on AlphaFold-style scientific discovery—on AI as an instrument of fundamental knowledge creation rather than incremental efficiency extraction—grows ever more compelling as the true north. The Rambo Liu-hosted People's Park Talks AI has consistently argued that emergent capabilities deserve compute allocation proportional to their civilizational returns; the present moment, where commercial LLM deployment and regulatory brinkmanship consume the industry's attention and resources, only sharpens that critique. The real question is whether the field can extricate itself from this double bind—where neither proprietary moats nor open-source abundance proves economically or politically stable—to pursue the kind of breakthroughs that genuinely expand the frontier of what AI can discover about the world.