DeepSeek's Fallout For AI Companies, OpenAI’s Path Forward, Siri Somehow Got Worse

Episode Summary

Executive Summary: The episode argues DeepSeek fundamentally shifted AI economics by proving powerful reasoning models can be built and run far more cheaply, weakening the case for margins in model APIs while strengthening the application layer. The hosts debate winners and losers—OpenAI, Anthropic, NVIDIA, Meta, Amazon, and Apple—concluding open source and products matter more than raw compute, while Apple’s Siri and AI strategy look increasingly weak.

Main Topics: DeepSeek rewrites AI economics (Priority: 5/5): The hosts frame DeepSeek as a turning point that makes efficient, low-cost model training and inference the new baseline, undermining assumptions that ever-larger model spend guarantees victory. OpenAI’s strategic position after DeepSeek (Priority: 5/5): They argue OpenAI’s true moat is ChatGPT and other applications, not the API/model business, but worry the company’s messaging and product execution have looked defensive and flat-footed. Anthropic and the closed-model business model (Priority: 4/5): Anthropic is seen as more exposed than OpenAI because it has bet heavily on selling model/API access; Dario Amodei’s response is viewed as partly self-serving but intellectually serious. NVIDIA, compute, and the weakening of the 'more compute wins' thesis (Priority: 5/5): The conversation questions whether NVIDIA’s dominance is as durable as investors assumed, since DeepSeek suggests innovation can compensate for less advanced chips and lower compute budgets. Open source vs. proprietary AI (Priority: 5/5): The hosts repeatedly argue open source is gaining momentum, with DeepSeek’s release and Meta’s open-source strategy suggesting that the next wave of progress may accrue to ecosystems, not closed labs. Apple Intelligence and Siri’s decline (Priority: 4/5): A separate segment criticizes Apple for making Siri worse with Apple Intelligence, arguing the company rushed AI features for market reasons rather than user value. Jevons paradox and the app-layer opportunity (Priority: 4/5): Cheaper AI is expected to increase total usage and unlock more startups and applications, even if model-layer profits compress.

Key Arguments: AI model margins are likely collapsing because reasoning and inference can be done much more cheaply than previously assumed. OpenAI’s economic future is tied more to ChatGPT and product distribution than to selling APIs or raw model access. Anthropic is more vulnerable than OpenAI because its business is more dependent on API/model monetization. DeepSeek’s success weakens the belief that compute scale alone determines winners; product quality, distribution, and efficiency matter more. Open source progress can spread quickly across labs and startups, making closed proprietary models less defensible. NVIDIA remains strong, but the certainty behind its “infinite compute demand” narrative has been reduced. Apple’s AI rollout appears reactive and consumer-unhelpful, especially given Siri’s poor performance and falling iPhone sales. Lower AI costs should spur more experimentation and a larger application market, benefiting builders even if incumbents lose margin.

Data Points: OpenAI ChatGPT weekly users: 100 million to 300 million - Used to illustrate rapid growth in ChatGPT adoption over roughly a year. DeepSeek R1 API cost vs OpenAI o1: $2.19 per million tokens vs about $60 per million tokens - Example from a developer comparing comparable reasoning model usage costs. Alternative API provider cost for DeepSeek R1: around $10 per million tokens - Mentioned as a more stable provider option still far cheaper than OpenAI. OpenAI projected revenue mix: about 80-85% from applications/consumer subscriptions - Referenced as leaked financial projections suggesting OpenAI’s business was always supposed to be app-led. ChatGPT app revenue since launch: $529 million - Cited as OpenAI’s App Store revenue to show the consumer app is a meaningful business. AI app market size estimate: $2 billion in 2024 - Appfigures estimate cited to show an emerging AI consumer app market. OpenAI fundraising target: $40 billion - Reported talks for a new round amid DeepSeek-related uncertainty. OpenAI valuation in new round: up to $300 billion - The figure discussed during the fundraising segment. Previous OpenAI round: $6.6 billion at a $157 billion valuation - Late-2024 round used as comparison to show how quickly the valuation is rising. Potential SoftBank contribution: $15-25 billion - Mentioned as part of the planned $40 billion raise. Public estimate of DeepSeek training cost: $5-6 million - Discussed as an incomplete or misleading figure, likely only covering a final training run or inference-related costs. DeepSeek app-store performance: Top app store ranking - Cited as a major signal of consumer traction and a reason OpenAI felt threatened. Apple iPhone sales: Down year-over-year in Q4 - Used to support the claim that Apple’s hardware momentum is weakening. Siri accuracy example: 34% - John Gruber’s cited test of Siri answering Super Bowl-related questions correctly. NVIDIA market capitalization: $3 trillion - Used to explain the scale of investor exposure and why the sell-off mattered so much. DeepSeek model deployment: Available on AWS - Mentioned as evidence that open-source models are already spreading into mainstream cloud ecosystems.

Pivotal Quotes: "the biggest takeaway from this week is that if open AI thought it was running an API business or a model building business, before this week, right now, there's clarification. It's all about the applications." — Host: Summarizing the new market view of OpenAI after DeepSeek. "I think that's the central question for the entire technology industry." — Ranjan Roy: On whether more compute still guarantees winning in AI. "Siri is Super Dumb and Getting Dumber." — John Gruber: Title of the Daring Fireball post discussed during the Apple segment.

Implications: AI value is shifting from model exclusivity toward cheap inference, applications, and distribution. Open source may accelerate startup creation, while NVIDIA, Anthropic, and Apple face new pressure to prove durable moats.