Claim 3.1

Inference cost per unit of capability is falling rapidly.

EconomicsClean to resolveEconomicsThe stack

◷ Share view, as of June 2026: confidence, rationales and priors are hidden. The questions, stances, claims, tests and evidence stay visible.

would falsify this claim

Cost-per-token-at-fixed-quality stops declining.

Evidence

56 support4 contradict28 neutral

Supports· weight medium· DeepSeek Nears $14 Billion Funding Round, Double Its Original Target - Seoul Economic Daily (en.sedaily.com)· via DeepSeek doubles funding target to $14B on profitability and IPO preparations· Market & Valuation
Supports· weight medium· State of AI Report 2026 - FINAL (stateof.ai)· via State of AI Report 2026 benchmarks frontier labs, agent monetization, and safety incidents across 12 months· Technical Capability
Neutral· weight medium· Meta releases its first frontier model, Muse Wonder - TechCrunch (techcrunch.com)· via Meta launches Muse Wonder frontier models with API pricing and benchmark parity claims· Market & Valuation
Supports· weight medium· Behind the Term Sheet: Moonshot AI (cathaycapital.com)· via Moonshot AI's K2.5 achieves frontier coding and agentic benchmarks at one-fifth the inference cost of leading closed models· Technical Capability
Neutral· weight medium· Introducing Muse Spark 1.1 - AI at Meta (ai.meta.com)· via Meta launches Muse Spark 1.1, multimodal agentic model with 1M-token context and structured coding improvements· Market & Valuation
Supports· weight medium· Microsoft unveils new AI models lessen reliance on OpenAI ... (cnbc.com)· via Microsoft launches proprietary MAI-Code-1-Flash and MAI-Thinking-1 models to reduce OpenAI dependency and lower inference costs for developers· Market & Valuation
Neutral· weight medium· NVIDIA and Microsoft Reinvent Windows PCs for the Age ... (investor.nvidia.com)· via NVIDIA RTX Spark: 1-petaflop on-device superchip for Windows local agents announced for fall 2026 deployment· Technical Capability
Neutral· weight medium· NVIDIA Launches Cosmos 3, the Open Frontier Foundation ... (hpcwire.com)· via NVIDIA launches Cosmos 3 open world foundation model for physical AI with mixture-of-transformers architecture· Technical Capability
Supports· weight medium· DeepSeek just dropped two insanely powerful AI models that rival GPT-5 and they're totally free | VentureBeat (venturebeat.com)· via DeepSeek V3.2 released open-source under MIT license with frontier-parity benchmark scores and 70% inference cost reduction· Technical Capability
Neutral· weight medium· Introducing GPT‑5 for developers (openai.com)· via OpenAI releases GPT-5 API with agentic capabilities: 74.9% on SWE-bench Verified, native tool chaining for developers· Technical Capability
Neutral· weight medium· Nvidia's stock market value hits $4 trillion on AI dominance (reuters.com)· via Nvidia reaches $4 trillion market cap on AI infrastructure dominance; Q2 guidance $45B revenue, up 22% YTD· Market & Valuation
Neutral· weight medium· Introducing OpenAI o3 and o4-mini (openai.com)· via OpenAI releases o3 and o4-mini reasoning models with tool-integrated agentic capabilities· Technical Capability
Neutral· weight medium· NVIDIA Announces Financial Results for Fourth Quarter ... (nvidianews.nvidia.com)· via NVIDIA Fiscal 2025: $130.5B revenue, +114% YoY; Q1 FY26 outlook $43B with gross margin compression· Market & Valuation
Neutral· weight medium· Introducing Gemini 2.0: our new AI model for the agentic era (blog.google)· via Google releases Gemini 2.0 Flash with native multimodal output and tool-use capabilities, positioning for agentic applications· Technical Capability
Neutral· weight medium· Introducing ChatGPT Pro (openai.com)· via OpenAI launches ChatGPT Pro, $200/month tier with unlimited access to o1 and o1 pro mode (inference-time scaling)· Market & Valuation
Neutral· weight medium· Microsoft invests €4.3B to boost AI infrastructure and cloud capacity in Italy - Microsoft News Center Italy (news.microsoft.com)· via Microsoft commits €4.3B to Italian cloud and AI infrastructure over two years· Market & Valuation
Supports· weight medium· Open Source AI is the Path Forward (about.fb.com)· via Meta releases Llama 3.1 405B as frontier-parity open-weight model, claims cost efficiency 50 pct below GPT-4o inference· Market & Valuation
Supports· weight medium· Introducing Claude 3.5 Sonnet (anthropic.com)· via Claude 3.5 Sonnet outperforms prior flagship on benchmarks and agentic reasoning at mid-tier speed and cost· Technical Capability
Supports· weight medium· Introducing Apple's On-Device and Server Foundation ... (machinelearning.apple.com)· via Apple releases 3B on-device and larger server foundation models with quantization and adapter-based specialization, claims performance parity with 7B-70B competitors on human evaluation· Technical Capability
Supports· weight medium· Nvidia dominates the AI chip market, but there's rising ... (cnbc.com)· via Nvidia AI chip market share estimated 70-95%, but hyperscalers and startups scaling homegrown silicon alternatives· Market & Valuation
Neutral· weight medium· Intel forecast misses estimates; shares tumble | Reuters (reuters.com)· via Intel projects Gaudi 3 AI accelerator revenue exceeding $500M in 2024 as enterprise AI demand shifts from CPU-centric architecture· Market & Valuation
Supports· weight medium· Introducing Meta Llama 3: The most capable openly available ... (ai.meta.com)· via Meta releases Llama 3 open-weight models (8B, 70B) claiming state-of-the-art performance on standard and custom benchmarks, with 15 trillion token training dataset· Technical Capability
Supports· weight medium· NVIDIA Blackwell Platform Arrives to Power a New Era of ... (nvidianews.nvidia.com)· via NVIDIA Blackwell GPU announced with 25x inference cost and energy reduction vs. Hopper, backed by all major cloud providers and AI labs· Market & Valuation
Supports· weight medium· Introducing the next generation of Claude (anthropic.com)· via Anthropic launches Claude 3 family with benchmark improvements and near-instant inference speed, addresses hallucination and refusal rates· Technical Capability
Neutral· weight medium· Nvidia stock surges as revenue forecast tops estimates, AI demand continues | Reuters (reuters.com)· via Nvidia Q4 2024 data-center revenue beats 9.5% above consensus; gross margin guidance 77% midpoint· Market & Valuation
Supports· weight medium· Google launches its largest and 'most capable' AI model, Gemini (cnbc.com)· via Google launches Gemini, claims Gemini Ultra outperforms GPT-4 on benchmarks, begins API access December 13· Technical Capability
Neutral· weight medium· Microsoft reveals Maia AI processor and Cobalt Arm-based chip (cnbc.com)· via Microsoft announces Maia AI processor and Cobalt ARM chip; Maia under internal testing for ChatGPT and Copilot workloads, no production timeline disclosed· Market & Valuation
Supports· weight medium· OpenAI to make models cheaper, more powerful in its first developer conference | Reuters (reuters.com)· via OpenAI previews cost reduction and vision capabilities at developer conference, November 2023· Market & Valuation
Supports· weight medium· Mistral 7B (mistral.ai)· via Mistral 7B achieves 3x equivalent model-size efficiency on reasoning tasks versus Llama 2· Technical Capability
Supports· weight medium· Generative AI and Disinformation (bipartisanpolicy.org)· via Generative AI lowers barriers to large-scale disinformation production, reducing cost and operational complexity· Technical Capability
Supports· weight medium· OpenAI intros new generative text features while reducing pricing | TechCrunch (techcrunch.com)· via OpenAI cuts GPT-3.5-turbo pricing 25%, text-embedding-ada-002 pricing 75%; ships function-calling and expands context window· Market & Valuation
Supports· weight medium· Will Big Tech's Rally Last? | J.P. Morgan (jpmorgan.com)· via NVIDIA Q2 guidance raised $4B ahead of consensus on AI chip demand; Big Tech portfolio concentration in S&P 500 returns rises to 81%· Market & Valuation
Neutral· weight medium· NVIDIA Announces Financial Results for First Quarter Fiscal 2024 | NVIDIA Newsroom (nvidianews.nvidia.com)· via NVIDIA Q1 FY24 revenue at $7.19B, data center at record $4.28B; Q2 FY24 outlook $11.0B signals accelerating AI infrastructure demand· Market & Valuation
Supports· weight medium· Meta releases LLaMA, triggering the open-weight LLM ... (epoch-arc.com)· via Meta's LLaMA release (February 2023) seeded open-weight ecosystem, enabling cost-competitive model development at sub-billion-parameter scales· Technical Capability
Neutral· weight medium· Nvidia's A100 is the $10,000 chip powering the race for A.I. (cnbc.com)· via A100 GPU becomes critical bottleneck for frontier AI model deployment, infrastructure capex skyrockets· Market & Valuation
Neutral· weight medium· Reinventing search with a new AI-powered Microsoft Bing and Edge, your copilot for the web - The Official Microsoft Blog (blogs.microsoft.com)· via Microsoft launches AI-powered Bing and Edge with chat and content-generation features, powered by next-generation OpenAI model· Market & Valuation
Supports· weight medium· AI and American Workers | National Affairs (nationalaffairs.com)· via Code flywheel and falling AI inference costs replicate Jevons paradox, raising structural labor absorption questions· Labor & Knowledge Work
Supports· weight medium· Introducing Claude Opus 5.5 - Anthropic (anthropic.com)· via Anthropic releases Claude Opus 5.5 with 40% lower inference cost and higher agentic reliability· Market & Valuation
Supports· weight medium· OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes - TechCrunch (techcrunch.com)· via OpenAI reduces GPT-6 Sol and Luna API costs by 50 percent with improved factuality· Market & Valuation
Neutral· weight medium· Meta debuts Muse Spark 1.3 as personal agent work continues - axios.com (axios.com)· via Meta debuts Muse Spark 1.3 as personal agent work continues· Technical Capability
Neutral· weight medium· OpenAI puts Pro subscriptions on hold due to Astra demand - TechCrunch (techcrunch.com)· via OpenAI pauses Pro subscriptions due to unprecedented Astra demand· Market & Valuation
Supports· weight medium· Abacus.AI Smaug Models Cut Agent Costs 100x [2026] (shattered.io)· via Abacus.AI releases Smaug open-weight models targeting enterprise agent costs at 10-100x below frontier APIs· Market & Valuation
Neutral· weight medium· On-premise medical AI agents for reliable clinical decision-making (nature.com)· via Study on on-premise medical AI agents for clinical decision-making published in Nature Medicine· Technical Capability
Neutral· weight medium· Vals, backed by Andreessen Horowitz, is looking to become the gold standard for AI benchmarking - techcrunch.com (techcrunch.com)· via Vals AI raises $40M to become the gold standard for AI benchmarking· Market & Valuation
Supports· weight medium· A new kind of AI model from a ChatGPT inventor is thrilling developers - TechCrunch (techcrunch.com)· via TypeSafe AI releases Jev, a non-LLM probability model for automation, drawing developer demand· Market & Valuation
Contradicts· weight medium· Rental Rates for Older AI Chips Jump 25% as Demand Surges - Seoul Economic Daily (en.sedaily.com)· via Rental rates for four-year-old H100 GPUs rise ~22-25% in Korea as compute demand spills over to older chips· Market & Valuation
Supports· weight medium· Anthropic releases new models, cuts agent costs - Axios (axios.com)· via Anthropic releases Fable 5.1, cuts agent costs, and reduces safety interventions· Market & Valuation
Supports· weight medium· Gemini 3.8 Flash Launches at $0.75/1M Input [2026] (tech-insider.org)· via Google Ships Gemini 3.8 Flash at $0.75/M Tokens, Fourth Flash-Tier Release in Four Months· Market & Valuation
Supports· weight medium· Research acceleration: The view inside OpenAI | OpenAI (openai.com)· via OpenAI reports research organization now runs 3.1 agent-workdays per human workday· Technical Capability
Neutral· weight medium· JPMorgan upgrades Meta after company’s recent AI product announcements - cnbc.com (cnbc.com)· via JPMorgan upgrades Meta on agentic traction but forecasts $65B to $75B annual free cash flow deficit through 2028· Market & Valuation
Supports· weight medium· Introducing Fugu Max and Fugu Ultra v2: Orchestrating the Pareto Frontier (sakana.ai)· via Sakana AI releases Fugu Max and Fugu Ultra v2 orchestration systems, claims improved cost-performance frontier and high benchmark scores without frontier models· Technical Capability
Supports· weight medium· China's Open-Weight Challenge to U.S. AI Leadership - CSIS (csis.org)· via China's Open-Weight Models Narrow the Gap with US Frontier Models· Geopolitics & Security
Supports· weight medium· Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear - TechCrunch (techcrunch.com)· via Salesforce launches Koa, an open-weight reasoning model post-trained on synthetic data to replace frontier API calls· Market & Valuation
Supports· weight medium· The Grid Doesn’t Care How Smart Your Model Is - The National Interest (nationalinterest.org)· via New research on intelligence per watt suggests efficiency, not frontier chips, may be the binding constraint on AI scale· Geopolitics & Security
Supports· weight medium· IDC - Tokenmaxxing Is Dead. AI Cost Governance Isn't (idc.com)· via Tokenmaxxing fades, but enterprise AI budget governance gap persists, IDC surveys show· Market & Valuation
Supports· weight medium· OpenAI Jalapeño Chip Beats Nvidia by 1.9x on Power [2026] (tech-insider.org)· via OpenAI publishes Jalapeño inference benchmarks showing 1.5x to 1.9x power efficiency gains over Nvidia Blackwell systems· Market & Valuation
Contradicts· weight medium· American AI Companies Can't Get Enough Chips (cnas.org)· via AI chip supply declared binding constraint on compute buildout as TSMC wafer capacity runs at or above 100% utilization· Market & Valuation
Supports· weight medium· OpenAI Enterprise Revenue Tops Consumer for First Time: $40 Billion ARR Two Quarters Early (techtimes.com)· via OpenAI enterprise revenue surpasses consumer for first time, ARR reaches $40B two quarters ahead of internal forecast· Market & Valuation
Supports· weight medium· China's Moonshot, Z.AI, and DeepSeek are challenging U.S. AI labs—and beating them on cost (fortune.com)· via Chinese frontier models reach near-parity with US labs on capability while undercutting on price by up to 98%, driving measurable enterprise adoption· Market & Valuation
Supports· weight medium· Six investment firms including Apollo and Blackstone partner with Nvidia to line up $500 billion in AI infrastructure financing for Nvidia customers (fortune.com)· via Enterprise AI cost overruns prompt usage caps and model downgrades as 2026 budgets exceed projections without clear ROI· Market & Valuation
Supports· weight medium· Writer launches Palmyra X6, built on Z.ai's GLM-5.2, promising up to 50 percent cost reduction for enterprise AI deployments (techcrunch.com)· via Writer launches Palmyra X6 on open-weight GLM-5.2 base, claims up to 50% cost reduction via model and harness optimization· Market & Valuation
Supports· weight medium· NVIDIA CORP - Form ARS - FY2026 (sec.gov)· via NVIDIA FY2026 annual report: revenue $216B (+65% YoY), Data Center $194B (+68%), operating cash flow $103B, with $1T+ cumulative Blackwell/Rubin revenue visibility through 2027· Market & Valuation
Supports· weight medium· The 2026 State of B2B SaaS and AI Monetization Report (growthunhinged.com)· via Survey of 230 B2B software firms finds AI gross margins at 50%, pricing models shifting toward hybrid and usage-based structures· Market & Valuation
Supports· weight medium· Poolside Drops Laguna S 2.1, an Open-Weight Coding Model That Beats Rivals 10x Its Size (venturebeat.com)· via Poolside releases Laguna S 2.1, a 118B MoE open-weight coding model outperforming rivals up to 1.6T parameters on agentic benchmarks· Technical Capability
Supports· weight medium· Microsoft Releases MAI-Image-2.5-Pro and MAI-Voice-2-Flash, In-House Models It Claims Cut Costs Up to 89% Versus OpenAI (venturebeat.com)· via Microsoft deploys in-house MAI models across production products, claims up to 89% GPU cost reduction versus OpenAI equivalents· Market & Valuation
Supports· weight medium· Anthropic Launches Opus 5, a Cheaper and Less Restrictive Frontier Model That Outperforms Fable 5 on Key Benchmarks (techcrunch.com)· via Anthropic launches Opus 5, outperforming Fable 5 on key benchmarks at lower cost and with lighter safety restrictions· Technical Capability
Neutral· weight medium· Large Language Models, Small Labor Market Effects· via Large-scale Danish employer-employee study finds near-zero earnings and hours effects from AI chatbot adoption across 11 white-collar occupations through mid-2024· Market & Valuation
Supports· weight medium· AI, Productivity, and Labor Markets: A Review of the Empirical Evidence (laweconcenter.org)· via ICLE review of empirical AI literature finds firm-level productivity gains of 15–55% but no aggregate labor displacement through 2025· Labor & Knowledge Work
Neutral· weight medium· The 2026 AI Index Report | Stanford HAI (hai.stanford.edu)· via Stanford HAI 2026 AI Index: Benchmark acceleration, US-China gap closure, and agentic leap headline annual data review· Technical Capability
Supports· weight medium· Meta publicly launches Muse Spark 1.1, a multimodal agentic coding model priced at $1.25 per million input tokens, entering direct competition with OpenAI and Anthropic (techcrunch.com)· via Meta launches Muse Spark 1.1 agentic coding model at $1.25/M input tokens, entering direct price competition with OpenAI and Anthropic· Market & Valuation
Neutral· weight medium· OpenAI releases full-duplex voice models GPT-Live-1 and GPT-Live-1 mini, enabling simultaneous listening and speaking with live translation (techcrunch.com)· via OpenAI releases full-duplex voice models GPT-Live-1 and GPT-Live-1 mini, replacing Advanced Voice Mode in ChatGPT· Technical Capability
Supports· weight medium· AI infrastructure spend in 2026 projected at $1.5 trillion, requiring $3 trillion in revenue to justify, as Anthropic reportedly hits $60 billion ARR (techcrunch.com)· via 2026 AI infrastructure spend estimated at $1.5T, implying $3T revenue requirement to justify investment· Market & Valuation
Supports· weight medium· OpenAI publicly launches GPT-5.6 family (Sol, Terra, Luna) after staggered rollout at US government request, claiming top coding and cybersecurity performance (techcrunch.com)· via OpenAI launches GPT-5.6 family (Sol, Terra, Luna) with claimed 54% coding token efficiency gain and top benchmark scores against Anthropic· Market & Valuation
Supports· weight medium· China's Moonshot AI releases Kimi K3, the largest open-weight model ever built, with 2.8 trillion parameters rivaling top US frontier models (venturebeat.com)· via Moonshot AI releases Kimi K3, a 2.8T-parameter open-weight model benchmarking near US frontier proprietary systems· Technical Capability
Supports· weight medium· The Week's 10 Biggest Funding Rounds: AI Drives Another Spree Of Megadeals (news.crunchbase.com)· via AI inference and infrastructure startups raise over $2.5B in a single week, led by Baseten $1.5B Series F at $13B valuation· Market & Valuation
Supports· weight medium· AI inference costs set to plunge: Gartner (ciodive.com)· via Gartner projects 90%+ inference cost drop by 2030 but forecasts enterprises absorb no net savings as agentic workloads raise per-task token spend· Market & Valuation
Supports· weight medium· AI Agents Forecast to Boost Tech Cash Flow as Usage Soars (goldmansachs.com)· via Goldman Sachs Research forecasts 24x token consumption growth by 2030, projects hyperscaler gross margin inflection from falling inference costs· Market & Valuation
Supports· weight medium· Your AI Feature Is Quietly Destroying Your Gross Margin (thesaascfo.com)· via ICONIQ 2026 survey: AI product builders expect blended gross margins of ~52%, versus 70–80% for traditional SaaS· Market & Valuation
Neutral· weight low· Large Language Models, Small Labor Market Effects
Neutral· weight medium· Memory Wall Bottleneck: AI Compute Sparks Memory Supercycle (trendforce.com)· via HBM and DDR5 demand surge drives memory supercycle; DRAM prices projected to rise 70%+ in 2026· Market & Valuation
Supports· weight medium· The Price of Progress: Price Performance and the Future of AI (arxiv.org)· via Benchmark price-performance falling 5–10× per year, but frontier model inference costs rising 3–18× per year as compute demands grow· Market & Valuation
Neutral· weight medium· The Great GPU Shortage – Rental Capacity – Launching our H100 1 Year Rental Price Index (newsletter.semianalysis.com)· via H100 1-year GPU rental prices rise ~40% in five months as compute demand outpaces supply across all tenors· Market & Valuation
Supports· weight medium· DeepSeek unveils V4 model, with rock-bottom prices and close integration with Huawei's chips (fortune.com)· via DeepSeek V4 launches at $3.48/M output tokens (~89% below OpenAI GPT-5.4 pricing) with native Huawei Ascend support· Market & Valuation
Contradicts· weight medium· Memory Chip Shortage 2026: HBM Takes 23% of DRAM Wafers (tech-insider.org)· via HBM wafer allocation hits 23% of DRAM production as data centers claim 70% of global memory output in 2026· Market & Valuation
Supports· weight medium· Trump Lifted the AI Chip Ban on China, Clearing Nvidia and AMD to Resume Sales: Now What? (builtin.com)· via US resumes H20/H200/MI308 AI chip sales to China under 15% revenue-sharing arrangement; policy oscillated four times in 13 months· Geopolitics & Security
Contradicts· weight medium· AI Prices Are Going Up, Up, Up - And What This Means For Enterprise AI (joshbersin.com)· via Enterprise AI pricing pressures mount as annualized hyperscaler + ecosystem capex approaches $1T, forcing revenue recovery through higher end-user costs· Market & Valuation
Supports· weight medium· H100 Rental Prices Compared: $1.49-$6.98/hr Across 15+ Cloud Providers (2026) (intuitionlabs.ai)· via H100 on-demand cloud rental rates fall to $1.49–$6.98/GPU-hr across 15+ providers by mid-2025, down ~44–73% from 2023 launch highs· Market & Valuation
Supports· weight medium· LLM inference prices have fallen rapidly but unequally across tasks (epoch.ai)· via LLM inference cost per capability milestone falling at 9x–900x per year, median ~50x, with acceleration post-January 2024· Market & Valuation

Signals on the board

Oct 14, 2026DeepSeek doubles funding target to $14B on profitability and IPO preparations Oct 12, 2026Meta launches Muse Wonder frontier models with API pricing and benchmark parity claims Oct 8, 2026State of AI Report 2026 benchmarks frontier labs, agent monetization, and safety incidents across 12 months Sep 22, 2026OpenAI reduces GPT-6 Sol and Luna API costs by 50 percent with improved factuality Sep 22, 2026Anthropic releases Claude Opus 5.5 with 40% lower inference cost and higher agentic reliability Sep 21, 2026Code flywheel and falling AI inference costs replicate Jevons paradox, raising structural labor absorption questions Sep 21, 2026Rental rates for four-year-old H100 GPUs rise ~22-25% in Korea as compute demand spills over to older chips Sep 19, 2026Vals AI raises $40M to become the gold standard for AI benchmarking Sep 18, 2026TypeSafe AI releases Jev, a non-LLM probability model for automation, drawing developer demand Sep 15, 2026Salesforce launches Koa, an open-weight reasoning model post-trained on synthetic data to replace frontier API calls

Research bearing on this · advisory

2026-09-10LOCUS: Task-Aware Low-Rank Post-Training for Token-Efficient Language Generation 2026-09-10COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization 2026-09-10FlexComp: One Model for Every Ratio in Context Compression 2026-09-10Building py-kvcache: A Performance Characterization of External KV Caching for vLLM with NVMe SSDs 2026-09-10REVA: Reusable Evidence View Aggregation for Context-Efficient RAG Serving

Bears on

supportsQ3-S3A Margins are structurally improving; the model is viable; we're early.