Unit cost of intelligence will drop at a rate far exceeding Moore's Law, driving explosive demand growth that filters out to edge devices and localized small models.

Unit cost of intelligence will drop at a rate far exceeding Moore's Law, driving explosive demand growth that filters out to edge devices and localized small models. | Next API token-pricing drop from frontier labs

Key catalyst: Next API token-pricing drop from frontier labs

Watch events: OpenAI / Anthropic / Google API token pricing; Epoch AI cost-per-FLOP tracker

OpenAI API token pricing fell ~40x 2023-2026 per Altman; DeepSeek-V3 / Kimi K2 / Claude Haiku 4.5 delivered frontier capability at 10-30x lower cost per token 2025-2026.

No propagation data yet. Run inference/.venv/bin/python scripts/ops/run_loopy_belief_propagation.py on the droplet, or wait for the Sunday 02:00 UTC weekly cron.

{
  "nia": false,
  "qty": ">Moore pace",
  "mode": "FORECAST",
  "role": "Cited-VC",
  "context": "Pairs with Altman's 40x/yr hyperdeflation call (235_014) and Huang tokens-per-watt KPI (CMQ_024). Directly downstream consequence of the oversupply thesis (INF_042).",
  "to_year": 2031,
  "conv_cues": "exceeds Moore's Law framing; universally-diffusing prediction",
  "direction": "DOWN",
  "from_year": 2026,
  "timeframe": "2026-2031",
  "conv_level": "HIGH",
  "milestones": [
    {
      "kind": "llm_pre_event",
      "label": "DeepSeek-driven 80%+ price drop across frontier LLMs (2024-2026)",
      "notes": "HIT — direct measurement of cost-of-intelligence trajectory exceeding Moore's Law.",
      "source": "TLDL/BenchLM 2026: LLM API prices dropped 80% across the board 2024->2026; GPT-4-class went from $30/M to $2-3/M (10x drop)",
      "status": "hit",
      "weight": 0.4,
      "ordinal": -10,
      "source_id": null,
      "confidence": 0.99,
      "source_url": "https://benchlm.ai/llm-pricing",
      "expected_date": "2026-04-30",
      "observed_date": "2026-04-15",
      "research_origin": "deep_research",
      "measurement_criterion": "Industry pricing trackers (PriceperToken, BenchLM, Iternal) document that headline frontier-LLM input-token API prices have dropped >=80% from early-2024 levels"
    },
    {
      "kind": "llm_pre_event",
      "label": "Anthropic Opus pricing cut 67%",
      "source": "Anthropic dropped Opus pricing 67% and expanded context window to 1M tokens (cited in 2026 pricing analyses)",
      "status": "hit",
      "weight": 0.4,
      "ordinal": -9,
      "source_id": null,
      "confidence": 0.95,
      "source_url": "https://intuitionlabs.ai/articles/ai-api-pricing-comparison-grok-gemini-openai-claude",
      "expected_date": "2026-04-30",
      "observed_date": "2026-03-31",
      "research_origin": "deep_research",
      "measurement_criterion": "Anthropic announces >=50% reduction in Opus or equivalent flagship model API pricing alongside context-window expansion"
    },
    {
      "kind": "llm_pre_event",
      "label": "Apple Intelligence runs 3B model on >=500M devices",
      "notes": "Confirms 'filters out to edge devices and localized small models' — direct mechanism for Andreessen claim.",
      "source": "Apple Intelligence 3B on-device model on iPhone 15 Pro+; over 2B smartphones run local SLMs (2026 edge-AI literature)",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -8,
      "source_id": null,
      "confidence": 0.95,
      "source_url": "https://semiengineering.com/the-on-device-llm-revolution/",
      "expected_date": "2026-06-30",
      "research_origin": "deep_research",
      "measurement_criterion": "Apple discloses Apple Intelligence (3B-param on-device model) deployed on >=500M iPhone 15 Pro / iPhone 16+ active devices"
    },
    {
      "kind": "quartile_checkpoint",
      "label": "Q1 window check-in (25%)",
      "status": "pending",
      "weight": 0.05,
      "ordinal": -7,
      "source_id": null,
      "expected_date": "2027-02-15",
      "observed_date": null
    },
    {
      "kind": "llm_pre_event",
      "label": "Sub-1B parameter SLMs match GPT-3.5 performance benchmarks",
      "source": "Edge AI Vision Alliance 2026: distilled small models outperform much larger base models on math/reasoning",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -6,
      "source_id": null,
      "confidence": 0.85,
      "source_url": "https://www.edge-ai-vision.com/2026/01/on-device-llms-in-2026-what-changed-what-matters-whats-next/",
      "expected_date": "2027-03-17",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-12-31",
        "from": "2026-06-01"
      },
      "measurement_criterion": "Public benchmarks (MMLU, HellaSwag, Big-Bench) show sub-1B-param model (Gemma 3 270M, SmolLM2, Phi-mini) matching or exceeding GPT-3.5 (175B) baseline"
    },
    {
      "kind": "scenario_signal",

... (truncated)