Where the math is defensible.
Long-form research on live enterprise decisions. Publication is selective. Every number traces to a named source. No takes without evidence.
Compute behind a fence: US AI export controls in 2026
Four years of BIS rules have built a tiered global compute regime. The October 2022 baseline, the October 2023 patch, the December 2024 HBM and tooling rules, and the January 2025 AI Diffusion Framework now shape every chip flow worth tracking. DeepSeek R1 forced the harder question. Are the controls working.
United States chip export controls have moved from a narrow performance ceiling into a full architecture for who gets to compute at frontier scale. The October 7, 2022 BIS rule set the first thresholds. October 17, 2023 closed the H800 and A800 workaround. December 2, 2024 added high bandwidth memory and 24 wafer fabrication tools to the ...
Frontier AI training cost trajectory 2026: the run rate, the deal stack, and the power-bound horizon
Frontier pretraining budgets crossed the half billion mark in 2025 and are heading toward one to three billion dollars per model by 2027, with cluster power, not GPUs, now the binding constraint on the next order of magnitude.
Frontier model training compute scaled at roughly 4x to 5x per year between 2018 and 2024, sat near 10x per year for the leading lab releases, and now confronts a deceleration driven by power, capital, and data, not by silicon. Epoch AI puts GPT-4 near 2e25 FLOPs at a roughly 80 million dollar training cost, Claude 3.5 Sonnet near 3e25, a...
Generative Video AI in 2026: Compute Economics, Hollywood Disruption, and the Copyright Reckoning
Frontier video models now generate 1080p clips in single-digit seconds, training runs cleared USD 100 million per system, and inference still costs cents per output second. The constraint is no longer fidelity but rights, latency, and where the marginal dollar of creative budget lands.
Generative video moved from research demo to production tool over fifteen months. OpenAI shipped Sora 2 in December 2024 with 8 second 1080p clips, Google Veo 2 and Veo 3 added native vertical and longer durations, Runway Gen-4 solved character consistency, and Chinese entrants Kling 2.0 from Kuaishou and Hailuo from MiniMax forced global...
Quantum Computing Economics in 2026: Roadmaps, Modalities, and the Post-Quantum Migration
IBM, Google, IonQ, Quantinuum, Atom Computing, PsiQuantum, and Microsoft are converging on fault tolerance from four physical platforms. Quantum equities have repriced sharply, NIST has finalized its post-quantum cryptography standards, and national programs have moved past 20 billion dollars of cumulative public commitment.
Quantum computing in 2026 is four parallel scaling experiments on transmon, trapped-ion, neutral-atom, and photonic platforms, judged against a shared bar set by the surface code. IBM ships Heron R2 at 156 qubits and targets Kookaburra in 2026 and Blue Jay at 4158 qubits. Google's Willow showed below-threshold operation in late 2024. IonQ...
UAE G42, sovereign AI ambitions, and the US China tech triangulation through 2026
G42 spent 2024 and 2025 converting Abu Dhabi political capital into US compute access, Chinese hardware divorce, and a balance sheet position inside the largest AI deals on the planet. Microsoft put 1.5 billion dollars on the cap table in April 2024, the Bureau of Industry and Security wrote a Diffusion framework in January 2025 that named the UAE explicitly, and Stargate UAE pushed five gigawatts of campus capacity to Abu Dhabi. The story is no longer whether the Gulf builds sovereign AI. The story is whose chips, whose models, and whose security guarantees the buildout runs on.
G42 was incorporated in Abu Dhabi in 2018 as a Mubadala adjacent holding company chaired by Sheikh Tahnoon bin Zayed Al Nahyan, the UAE National Security Advisor and brother of the President. Through 2023 the group accumulated stakes in TikTok parent ByteDance, Huawei surveillance, BGI Genomics sequencing partnerships, and a Wuxi cloud jo...