🧠 Food for thought

Implications, context, and why it matters.

Octen joins a crowded race against established players

• Legacy search APIs were built for human web searches. They can lag in AI use cases, and one provider charges up to $200 per thousand queries 1.
• That gap helped AI-native infrastructure take off. Perplexity, an AI search company, says its system handles 200 million queries a day and its index covers more than 200 billion unique URLs 1.
• Latency now sets the bar. Perplexity puts its median response at 358 milliseconds (ms) 1.
• Octen targets faster replies. It says its Web Search API averages 99 ms and supports more than 1 million queries per second (QPS) 2.

AI-native search drives a trillion-dollar infrastructure shift

• Perplexity expects AI agents to send “orders of magnitudes more requests” than legacy search engines. It says traditional systems topped out around 10^10 queries per day 1.
• Rising demand is pushing a new hardware buildout. AI data center capital expenditures (capex) are projected to hit $1 trillion by 2028 3.
• Pressure for lower delays is also speeding up consolidation. NVIDIA struck a roughly $20 billion licensing-and-acqui-hire deal with Groq, a company that builds AI accelerator chips, for its LPU (Language Processing Unit) technology and engineering team 3.
• Octen wants to rebuild the search layer as low-latency infrastructure for AI models rather than people, and it aims to ride this shift 2.