World Models in 2026: Genie 3 at 24 fps, Cosmos 3 at 64B, V-JEPA 2 Planning in 16 Seconds, and an $8.2B Exit to AMD
What a world model is in 2026 and what actually shipped: Genie 3 at 720p and 20 to 24 fps for a few minutes behind a $249.99 subscription, World Labs' Marble and Atlas and the $8.2 billion all-stock sale to AMD, NVIDIA Cosmos 3 at 64B, 16B and 4B under OpenMDW 1.1, V-JEPA 2 planning a robot action in 16 seconds against Cosmos's four minutes, LeCun's AMI Labs raising $1.03 billion at $3.5 billion, the Physics-IQ ceiling of 29.5 out of 100, what inference costs, and a V-JEPA 2 embedding script I ran on a laptop CPU.
14 minOct 5, 2026Model Collapse in 2026: The Loop Nobody Runs, a 40% Synthetic Phi-4, a 49.9% AI Web, and 300T Human Tokens Left
What the model-collapse research tested versus the headline: Shumailov's Nature loop deletes real data every generation and takes OPT-125m from 20 to 28 perplexity, while accumulating data bounds the error; my 100-generation Gaussian experiment shows median variance falling from 0.98 to 0.36 under replacement and holding at 0.98 under accumulation; how Phi-4 (40% synthetic), Llama 3, DeepSeek-R1 (800k samples), Qwen3, Kimi K2 and Nemotron-CC-v2 (38%) actually use it; Epoch's 300T-token stock, Graphite's 49.9% AI web, Cloudflare's 38,065:1 crawl ratio and the $1.5B Bartz settlement.
15 minOct 5, 2026GPU Clouds in 2026: H100s at $2.59 an Hour, $104B of CoreWeave Backlog on $35B of Debt, and a Six-Year Bet on Silicon
The GPU rental economy from filings and price pages fetched on 5 October 2026: H100 hours that fell from about $8 in 2023 to a $1.70 trough in October 2025 and recovered to $2.59 on the transacted index while list prices rose 25 percent; CoreWeave's $2.58B quarter, $103.7B backlog, $35.1B of debt from 15 percent DDTLs to a 5.9 percent A3 facility, and Microsoft at 67 percent of 2025 revenue; Nebius, Lambda, Crusoe, Oracle's $664B RPO and NVIDIA's $105B of guarantees; who books servers over 5, 5.5 or 6 years and what Burry's $176B means; and an own-versus-rent break-even calculator.
13 minOct 5, 2026FinOps for AI in 2026: 98% Manage Token Spend, 3 in 4 Can't Prove the Value, GPUs Run at 5%, and FOCUS Gets a Token Column in 1.5
The AI cost ledger as of October 2026: State of FinOps 2026 (1,192 practitioners, 98 percent manage AI spend, up from 31 percent in 2024), the Tokenomics Foundation finding that three in four enterprises cannot prove AI outcomes to the CFO, a per-million-token price table for Anthropic, OpenAI, Google and DeepSeek with cache and batch multipliers, a 20-step agent loop at $1.58 uncached and $0.42 cached, the FOCUS 1.2 to 1.5 roadmap and who exports which version, LiteLLM and Cloudflare budgets, Cast AI's 5 percent GPU utilisation, and $0.0037 of model cost inside a $0.99 resolution.
13 minSep 9, 2026On-Device Inference in 2026: 160 Tokens a Second on a Laptop, a 2-Bit Model in Every iPhone, and the Bandwidth Arithmetic That Decides What Stays Local
What actually runs on the device now: Apple's 3B model trained at 2 bits and the WWDC26 framework that lets any model plug in, Gemini Nano inside AICore, Windows ML and the quiet replacement of Phi Silica, the NPU TOPS numbers and why they are the wrong metric, the small-model field from Gemma 4 E2B to Qwen3.5-2B to LFM2.5 with licences, measured decode speeds from Google's own tables and an independent phone benchmark, the runtimes (llama.cpp with multi-token prediction, LiteRT-LM, ExecuTorch 1.0, MLX), the memory-bandwidth arithmetic that predicts all of it, and the model-extraction attacks nobody has fixed. With commands to run one today.
14 minSep 9, 2026AI Power in 2026: 485 TWh, a 0.3 GW Stargate, a 116 GW Turbine Backlog, and 1,800 MW That Dropped Off the Grid in Seconds
The electricity numbers behind AI, kept honest: IEA's 485 TWh for 2025 and 950 by 2030, LBNL's 4.7 percent of US power heading to 12, what a prompt actually costs (0.24 Wh at Google, 30 times more in reasoning mode), which gigawatt campuses are energised versus announced (Stargate Abilene at 0.3 of 1.2 GW), where the power comes from (a 116 GW gas-turbine backlog with 2031 slots, nuclear restarts in 2027, one SMR construction permit), what the grid operators are doing about 233 GW queues and 1,800 MW load-loss events, rack power from 132 kW to 600 kW, and a script to measure your own job's energy.
14 minSep 9, 2026Agent-to-Agent Protocols in 2026: A2A Reached 1.0, Three ACPs Died or Merged, and MCP Learned to Do Most of the Same Job
What Google's Agent2Agent protocol actually specifies at v1.0, what changed when it moved into the Agentic AI Foundation in August, which vendors have shipped it versus announced it, how MCP's July 2026 revision reproduced A2A's task lifecycle almost state for state, the first real CVE on an A2A surface, what the production data says about how rare multi-hop agent systems are, and a working server and client on the 1.1 SDK with the breaking changes listed.
14 minSep 8, 2026Open Weights in 2026: Chinese Labs Ship the Frontier, the Licenses Grew Clauses, and Brussels Started Enforcing
DeepSeek V4, Kimi K3, GLM-5.3, Qwen3.8 and Xiaomi's MiMo are the open frontier now; Meta went closed and came back; the licences quietly acquired revenue thresholds; the EU AI Office got its enforcement powers on 2 August. What I found when I read every licence file, with the Hugging Face API calls that check them.
16 minSep 8, 2026Confidential Computing for AI in 2026: Apple Moved to Google's TEEs, a $1,000 Interposer Broke the Keys, and Firecracker Still Can't Do It
What a hardware enclave actually promises for an AI workload, what it costs (4 to 28% depending on what you measure), what broke it this year (TEE.fail, Battering RAM, WireTap), who runs inference in one (Apple, Google, Meta, a dozen startups) and who doesn't say (OpenAI, Anthropic). Plus the state of SEV-SNP, TDX and Arm CCA in the kernel, why my microVM platform cannot offer it yet, and the commands to verify an attestation yourself.
15 minSep 8, 2026Agent Observability in 2026: The Spans Are Standard, the Standard Isn't Stable, and Nobody Reads the Logs
OpenTelemetry's GenAI conventions moved to their own repo in June and still carry the Development badge; every vendor from Datadog to Grafana ingests them anyway. I instrumented an agent loop with the official Python library, show the exact spans it emits, and go through what the 2026 eval papers say actually catches failures: pass^k, log inspection, and a step budget.
13 minSep 7, 2026LLM Inference in 2026: Prefill Is a Different Machine Now, and the KV Cache Has a Disk Tier
vLLM 0.28, SGLang on 400,000 GPUs, NVIDIA Dynamo 1.0 and llm-d in the CNCF all converged on the same architecture: split prefill from decode, route on cached prefixes, page the KV cache to CPU and SSD, and run MoE experts wide across a rack. What changed this year, with the papers, the flags, and the prices.
15 minSep 3, 2026AI Agent Cloud: What the Term Means and Who Actually Has One
An AI agent cloud needs more than a sandbox: state, hosting, triggers, audit. I map 15 platforms against that stack and show where each one stops short.
10 minSep 2, 2026What Scale to Zero Cost Me to Build: 50s Sleeps, 14s Wakes
Hibernate took 50 seconds. Wake took 14, not the 1.3 I had reported. The measured price of building scale to zero on a Firecracker fleet, and every fix.
9 minSep 2, 2026Checkpoint Restore Implementation: What It Cost Me to Build
Analysts call checkpoint/restore table stakes by 2027. Here is the real bill: snapshot storage, userfaultfd lazy restore, and a 50s hibernate vs 1.2s wake.
9 min