LIVE · SAT, JUL 11, 2026 --:--:-- ET
Issue Nº 81 COST TOTAL $14786.63 ARTICLES TODAY 1 TOKENS TOTAL 9.46B
Podcast
aiexpert
Subscribe · $8/mo
Running the wire
Funding Ollama raises $65M Series B as local AI adoption hits 8.9M monthly developers Chips Colibrì proof-of-concept runs 1.5TB frontier model on 25GB RAM; hints at consumer-scale inference Chips Meta to produce custom Iris AI chip starting September; targets 14 gigawatts by 2027 Market Meta, SpaceXAI, OpenAI unleash price war: new models priced 60–90% cheaper Market China's power load hits record 1.518 TWh on July 10 amid AI data center surge Funding Keyfactor raises $1B+ led by Summit Partners for AI and post-quantum machine identity management Market China power demand from data centers projected to grow 300–500 TWh by 2030, 18% of total growth Breaking SambaNova CEO signals 2027 IPO after $1B Series F at $11B valuation Chips SK hynix & TetraMem demo memristor edge-AI chip, 21.3 TOPS/W efficiency for lightweight inference Market Reflection AI commits $6.3B through 2029 for SpaceX Colossus GB300 capacity; $150M/month sovereign inference spend Breaking OpenAI declares GPT-5.6 preferred model for Microsoft 365 Copilot; family launches with Sol/Terra/Luna variants Market SK Hynix raises $26.5B in largest foreign IPO in U.S. history; commits $8.6B to ASML Breaking LangChain launches OpenWiki Brains: proactive memory framework for agents across email, Notion, GitHub, web Funding Prime Intellect raises $130M Series A at $1B valuation to decentralize frontier AI training Research NVIDIA Nemotron 3 Ultra hits LangChain benchmark lead, cheaper than frontier closed models Chips Meta's Iris AI chip enters production September; targets 14 gigawatts capacity by 2027 Funding Keyfactor secures $1B+ growth investment for AI-era machine identity and post-quantum security Market AI market shifts from biggest models to cheaper, smarter systems and open-weight alternatives Market SK Hynix raises $26.5B in largest US IPO by foreign firm, riding AI memory boom Breaking Apple sues OpenAI alleging systematic trade secret theft for hardware development Funding Ollama raises $65M Series B as local AI adoption hits 8.9M monthly developers Chips Colibrì proof-of-concept runs 1.5TB frontier model on 25GB RAM; hints at consumer-scale inference Chips Meta to produce custom Iris AI chip starting September; targets 14 gigawatts by 2027 Market Meta, SpaceXAI, OpenAI unleash price war: new models priced 60–90% cheaper Market China's power load hits record 1.518 TWh on July 10 amid AI data center surge Funding Keyfactor raises $1B+ led by Summit Partners for AI and post-quantum machine identity management Market China power demand from data centers projected to grow 300–500 TWh by 2030, 18% of total growth Breaking SambaNova CEO signals 2027 IPO after $1B Series F at $11B valuation Chips SK hynix & TetraMem demo memristor edge-AI chip, 21.3 TOPS/W efficiency for lightweight inference Market Reflection AI commits $6.3B through 2029 for SpaceX Colossus GB300 capacity; $150M/month sovereign inference spend Breaking OpenAI declares GPT-5.6 preferred model for Microsoft 365 Copilot; family launches with Sol/Terra/Luna variants Market SK Hynix raises $26.5B in largest foreign IPO in U.S. history; commits $8.6B to ASML Breaking LangChain launches OpenWiki Brains: proactive memory framework for agents across email, Notion, GitHub, web Funding Prime Intellect raises $130M Series A at $1B valuation to decentralize frontier AI training Research NVIDIA Nemotron 3 Ultra hits LangChain benchmark lead, cheaper than frontier closed models Chips Meta's Iris AI chip enters production September; targets 14 gigawatts capacity by 2027 Funding Keyfactor secures $1B+ growth investment for AI-era machine identity and post-quantum security Market AI market shifts from biggest models to cheaper, smarter systems and open-weight alternatives Market SK Hynix raises $26.5B in largest US IPO by foreign firm, riding AI memory boom Breaking Apple sues OpenAI alleging systematic trade secret theft for hardware development
Market

Meta, SpaceXAI, OpenAI unleash price war: new models priced 60–90% cheaper

In a span of 24 hours, Meta, SpaceXAI (formerly xAI), and OpenAI released new frontier models with dramatically slashed pricing, signaling the end of the era when raw intelligence commanded premium margins. SpaceXAI launched Grok 4.5 on July 8 at $2–6 per million tokens (60–76% cheaper than Anthropic's Claude Opus 4.8). Meta followed with Muse Spark 1.1 on July 9 at $1.25–4.25 per million tokens (roughly one-tenth competitors' cost). OpenAI released GPT-5.6 Luna at $1–6 per million tokens, underscoring competitive pressure from the cheaper alternatives.

The pricing war reflects two structural shifts: first, frontier model performance is commoditizing—Artificial Analysis' benchmarks show margin narrowing between labs, with SpaceXAI's Grok 4.5 and Meta's Muse Spark 1.1 competitive on coding and agentic tasks despite lower capability scores. Second, agentic AI workloads are driving token consumption rates previously unseen; enterprises face potential 'token bill shock' as agents read large codebases, call tools, and iterate for minutes. Goldman Sachs forecasts 120 quadrillion monthly tokens by 2030, making per-token efficiency critical to procurement.

Meta's strategy explicitly targets developers with aggressive API pricing 25% of OpenAI and Anthropic's cost, betting that lower cost will overcome perception gaps. OpenAI's Luna tier responds directly, abandoning previous premium positioning. SpaceXAI, leveraging Cursor coding-tool data from its $60 billion acquisition, positioned Grok 4.5 as 'Opus-class but faster, more token-efficient and lower cost.' Meanwhile, Google remains silent—Gemini 3.5 Flash has fallen to 10th on benchmarks since May.

For purchasing architects, this resets procurement calculus: task-specific model selection (not 'best-overall' vendor lock) is now table stakes. Token budgets must account for agentic overhead; estimated cost-per-task, not per-token, is the relevant denominator for enterprise workloads. OpenAI and Anthropic face margin compression; expect bundled agent/agentic services and embedded use cases to become differentiation levers rather than raw model capability.

Sources

Get the signal before the noise.