AI economics and markets
What AI costs once you count properly: token cost of ownership, the pressure on software business models, and how each technology wave priced itself.
Research sweeps
2026-08-14 · deep
AI and the Labour Market, Signal, Noise and Structural Change
Evidence of AI-related labour-market change from November 2022–August 2026: employment, vacancies, wages and productivity; occupational and demographic distribution; job creation, displacement and work redesign; and competing economic interpretations using ONS, BLS, Eurostat, OECD, ILO and IMF evidence.
GPT-5.6-sol- financial
- frontier
- academic
- +3
2026-08-04 · deep
Apple Silicon Unified Memory as On-Premise AI Infrastructure (Jan 2025 – Aug 2026)
Whether Apple Silicon unified memory and the Neural Engine are becoming a credible substrate for how AI is run commercially and on consumer devices between January 2025 and August 2026: shipped capacity and bandwidth (M3 Ultra at 512GB, M4 and M5 Max) versus the rumoured 1.5TB M7 Ultra, comparison against NVIDIA H200, GB200 NVL72 and AMD MI355X on memory-bound serving, the orchestration and scheduling gap (MLX distributed, EXO, Thunderbolt 5 fabrics, Kubernetes and MDM on macOS, Metal kernel maturity versus CUDA), the on-device stack (Neural Engine, Core ML, the Foundation Models framework, Private Cloud Compute), fleet monitoring and data-centre economics, and how Apple is positioned to capture value from the AI race as compute substrate rather than as a frontier model lab.
Claude Fable 5- frontier
- academic
- tech
- +3
2026-07-15 · deep
Enterprise AI Transformation Programmes (2025–2026)
Enterprise AI transformation programmes from July 2025 to July 2026: reported success and failure rates and who measures them, delivery frameworks borrowed from and diverging from classic digital transformation, token cost economics and budgeting under consumption pricing, adoption strategy including leadership over-provisioning of access, and AI governance across AI-Ops, financial business cases, security and regulatory controls as they scale with sector risk tolerance, referencing MIT, McKinsey State of AI, DORA, ThoughtWorks Technology Radar, NIST AI RMF, ISO 42001, and the EU AI Act.
Claude Fable 5- financial
- tech
- academic
- +3
2026-06-20 · deep
Comparative LLM Usage Across Sectors
Comparative real-world usage of LLMs and adjacent AI technologies from June 2025 to June 2026: which models (GPT-5, Claude, Gemini, Llama, Mistral, DeepSeek, Qwen) dominate which sectors, how they are deployed (hosted API, Bedrock/Azure, self-hosted vLLM/Ollama, RAG, agents, fine-tuning), what workloads they serve, and how organisations measure, budget, and publicly report token cost and actual spend.
Claude Opus 4.8- financial
- frontier
- academic
- +3
2026-05-11 · deep
Compounding Waves - How Each Tech Era Built the Substrate, and the Skills, for the Next
The compounding economic logic of three successive technology waves from January 1995 to May 2026 - internet disintermediation of distribution, software-defined platforms and cloud infrastructure, and the current AI/agentic systems wave - examining the technical, economic and human-skills dependencies that make each wave a precondition for the next, the new categories of work each wave created, and whether the relationship is best understood as cumulative compounding or as externalised costs harvested by later layers.
Claude Opus 4.8- financial
- academic
- blogs
- +1
2026-04-19 · deep
Token Cost of Ownership
AI token pricing vs true total cost of ownership from January 2023 to 19 April 2026, with emphasis on 2025–2026 signals: lab subsidisation strategies, infrastructure economics (compute, energy, data centres, hardware, security, ops), how user-facing prices have evolved, and analyst and researcher projections for token cost trajectories through 2028.
Claude Opus 4.8- financial
- frontier
- academic
- +2
2026-04-15 · standard
The SaaS-pocalypse - AI Displacement, Overhiring Hangover, or Multiple Compression?
The 2026 SaaS sector stress: testing whether weak SaaS revenue growth and stock performance are driven by AI displacing knowledge-work jobs, post-ZIRP overhiring correction, compression of growth-era revenue multiples, or macro tech-capex slowdown - January 2026 through April 2026.
Claude Opus 4.8- financial
- academic
- vc
- +1
2026-04-13 · standard
Enterprise LLM Vendor Selection and Consumption Models
Enterprise LLM vendor selection and consumption patterns (April 2025–present): how companies choose between OpenAI, Anthropic, Google, hyperscaler-hosted model access, and direct API relationships; what decision metrics they use across availability, quality, price, governance, and SLAs; and how adoption differs by company size, workload criticality, and realtime versus offline use cases
Claude Opus 4.8- financial
- frontier
- academic
- +2
Explainers
- Research Explainer · West (2026)
Free models agree with the expensive verifier when it says yes, but miss most of its real failures
Eight free candidates, five local open-weight models and three cloud free-tier routes, were retro-graded against frontier verdicts on ten real stages of agent-written code. The best caught 77% of genuine failures. Five caught almost none, while agreeing with passing work up to 98% of the time.
- Research Explainer · Hu et al. (2025)
EPIC reuses KV caches across any prefix, by recomputing only a handful of tokens per chunk
Position-Independent Caching lets language models reuse document KV vectors regardless of what comes before them. EPIC's LegoLink algorithm fixes the resulting attention sink with O(kN) work instead of O(N²).
- Research Explainer · Mertens et al. (2026)
AI is not crashing over jobs in waves, it is rising as a tide across nearly all of them
Across 17,000 worker evaluations of more than 3,000 real labor-market tasks, frontier models improve broadly across task lengths, not in sudden bursts. By 2029 most text-based tasks could hit 80–95% success rates.
- Research Explainer · Xing (2026)
AI inference tokens are becoming a commodity, and someone has designed the futures contract
This paper argues that the tokens consumed by large language models share the economic properties of electricity and carbon credits, then proposes a complete standardized futures contract to let enterprises hedge their compute costs.
- Research Explainer · Litowitz (2026)
The token economy has a finite energy budget, but the real bottleneck is knowing which questions to ask
Litowitz, Polson and Sokolov treat the AI token as a physical quantity with measurable thermodynamic cost, then build a MacKay-style balance sheet showing that projected 2028 US infrastructure could supply 225,000 tokens per person per day, over 1,000× current usage. The binding constraint, they argue, is not compute but the human capacity to formulate questions worth answering.
- Research Explainer · Chen (2025)
Nightly GPU benchmarks reveal no single vendor wins everywhere, but cost per token tells the real story
SemiAnalysis's InferenceMAX is an open-source, nightly benchmark that tracks throughput, latency, TCO per million tokens, and tokens per megawatt across NVIDIA and AMD GPUs, exposing how fast inference software improves and where each chip actually leads.
- Research Explainer · Wu, Sun, Li, Welleck & Yang (2025)
Smaller models with smarter inference beat bigger models, and it's not even close
A 7-billion-parameter model paired with a novel tree search algorithm consistently outperforms a model five times its size on maths benchmarks, using half the compute. The trick is spending your budget on thinking harder, not on being bigger.
- Research Explainer · Prucs, Csutora, Antal & Marosi (2025)
Reasoning models hit a compute ceiling, but sparse architectures keep climbing
A benchmark study of 19 open-source LLMs finds that Mixture-of-Experts models consistently dominate the accuracy-per-FLOP frontier, while all architectures eventually reach a task-dependent saturation point where more thinking time stops helping.
- Research Explainer · Zheng et al. (2024)
Treat LLM calls like a program; then the cache starts doing real work
SGLang pairs a small Python DSL with a runtime that understands prompt structure, shared prefixes, and batching. That combination makes agent, reasoning, long-document, and vision workloads measurably faster, while also cutting the amount of glue code needed to build them.
- Research Explainer · Pan, Chodnekar, Roy & Wang (2025)
Running your own LLM can pay for itself in months; but only if you pick the right model size
A cost-benefit analysis of 54 deployment scenarios finds that small open-source models break even against commercial APIs in under three months on a $2,000 GPU, while large models can take years to justify their quarter-million-dollar hardware.