Every statistic, timeline, quote, and technical detail on Silicon Epoch is pulled from real-world documents. Below is our transparency bibliography.
Author/Publisher: Focus Taiwan
Foundry capacity projections outlining TSMC's 2nm nanosheet output CAGR and fab allocations in Kaohsiung Fab 22 and Hsinchu Fab 20.
Author/Publisher: TSMC Official Press Release
Confirms N2 volume production ceremony, GAA nanosheet transistor curves, and wafer price metrics.
Author/Publisher: Tech Insider
Detailed engineering analysis of the Vera Rubin R100 GPU, detailing its N3 process node, HBM4 integration, and transistor count.
Author/Publisher: Spheron Network
System-level technical specifications for the Vera Rubin platform, NVLink 6, and HBM4 memory bandwidth.
Author/Publisher: Intel Corporation
Official datasheet for Gaudi 3, detailing its 1,835 TFLOPS BF16 compute performance, HBM2e memory, and TDP specs.
Author/Publisher: Intel/Introl Technical Group
Deployment metrics, pricing estimates, and shipping targets for Gaudi 3 AI accelerators.
Author/Publisher: Arc Compute Blog
Analyzes HBM4 supply crunches, Vera Rubin NVL72 liquid-cooling requirements, and data center thermal scaling challenges.
Author/Publisher: Meta Intelligence
Teardown of DeepSeek V4's 1-trillion parameter Mixture-of-Experts (MoE) model and its auxiliary-loss-free load balancing.
Author/Publisher: LMSYS Chatbot Arena
Aggregated benchmark rankings comparing open weights like Llama 4, Qwen3-Coder, Kimi K2.5, and Mistral Small 4.
Author/Publisher: Anthropic Developer Documentation
Details the Claude 4/5 pricing tiers, input/output cost structures, and context windows.
Author/Publisher: Simon Willison's Weblog
Initial coverage of Anthropic's dual Fable 5 and Mythos 5 models, safety router redirections, and Stripe integration trials.
Author/Publisher: Ortem Tech
Detailed overview of Gemini 3.5 Flash and Gemini 3.1 Pro's multi-million token context window, multimodal token pricing, and System 2 thinking features.
Author/Publisher: Digital Applied
Tracks OpenRouter API traffic, Chinese lab model price cuts, and deployment of domestic hardware stacks.
Author/Publisher: OpenAI Developer Relations
Official launch document outlining OpenAI's flagship GPT-5.5, the GPT-5.5 Instant tier, its pricing structure ($5/$30 per million), and performance indices.
Author/Publisher: Anthropic Press
Highlights the release of Claude Opus 4.8 featuring enhanced agentic planning capabilities, a 1M token context, and its SWE-bench Pro dominance.
Author/Publisher: xAI Blog
Outlines xAI's flagship Grok 4.3 release, the Colossus 2 GPU clusters, and the terminal-based Grok Build agentic coding interface.
Author/Publisher: Mistral AI Team
Official release announcement for Mistral Large 3 and Mistral Small 4, detailing model pricing, multilingual benchmarks, and agentic integrations.
Author/Publisher: Coding Fleet Blog
Explains the SWE-bench Pro dataset (1,865 tasks across 41 repos), data contamination on SWE-bench Verified, and model scores.
Author/Publisher: Morph LLM Blog
Evaluates Claude Fable 5, Claude Opus 4.8, GPT-5.5, Gemini 3.5 Flash, and MiniMax M2.5 on code planning and execution.
Author/Publisher: Trieu Trinh, John Jumper, et al. (Google DeepMind)
Details the architecture of AlphaGeometry 2, achieving an 84% solve rate on IMO geometry problems from 2000–2024.
Author/Publisher: Nature News
Nature's coverage of AlphaProof and AlphaGeometry 2 solving 4 of 6 IMO problems to earn a silver-medal score.
Author/Publisher: Frontiers in Artificial Intelligence
Explains AlphaFold 3's Pairformer module, ligand-docking accuracy metrics, and structural limitations.
Author/Publisher: Google DeepMind GitHub
Official code repository and dataset release for Graph Networks for Materials Exploration (GNoME), mapping stable crystals.
Author/Publisher: MMCG Invest Post
Comprehensive report on Meta's global datacenter footprint, highlighting the 1 GW Ohio Prometheus and 5 GW Louisiana Hyperion campuses.
Author/Publisher: Constellation Energy
Press release detailing the 20-year Microsoft PPA to restart Three Mile Island Unit 1 (835 MW) as the Crane Clean Energy Center.
Author/Publisher: ABC News / AP News
Details the U.S. Department of Energy's $1B federal loan to Constellation to fast-track the Crane Center restart for Microsoft AI.
Author/Publisher: Google Official Blog
Coverage of Google's PPA to purchase 500 MW from a fleet of Kairos Small Modular Reactors, targeting Oak Ridge, TN.
Author/Publisher: Baxtel Data Center Registry
Technical profile of Amazon's datacenter campus, situated adjacent to Talen Energy's 1,920 MW Susquehanna plant in Pennsylvania.
Author/Publisher: American Nuclear Society (ANS)
Details on the 1,920 MW Amazon-Talen PPA through 2042 and the spring 2026 front-of-meter grid interconnection.
Author/Publisher: Bits & Chips Magazine
Reports ASML's shipment of the pre-production 0.55 NA tool to Samsung Hwaseong, following Intel and TSMC.
Author/Publisher: Wikipedia
Comprehensive history of U.S. export controls, including the October 2023 BIS rules and subsequent 2024–2026 revisions.
Author/Publisher: Introl Policy Review
Detailed overview of the January 15, 2026 BIS rule setting TPP and bandwidth limits, and ByteDance's $14B order pipeline.
Author/Publisher: Enkia Market Intelligence
Tracks SMIC's 7nm yields (20-40%), DUV multi-patterning restrictions, and 5nm pilot runs for Huawei Ascend.
Author/Publisher: SemiconductorX
Deep-dive report on Huawei's Ascend 910C and 950PR/DT architectures, custom memory stacks, and server cabinet integrations.
Author/Publisher: Gibson Dunn
Legal analysis of Governor Newsom's veto of the Safe and Secure Innovation for Frontier AI Models Act and its impact on open weights.
Author/Publisher: California LegiScan Portal
Full official text of SB 1047, outlining the $100M training compute cost and 10^26 FLOP thresholds.
Author/Publisher: Towards AI
Outlines Epoch AI's public human data depletion timeline, 5× overtraining multiples, and MIT crawler-blocking datasets.
Author/Publisher: OpenAI
Explains System 2 reasoning architectures, test-time token budgets, and reinforcement learning optimization routes.
Author/Publisher: Hugging Face Blog
Technical post explaining inference-time scaling, search tree generation, and DeepSeek-R1's GRPO verification pipelines.
Author/Publisher: SyncSoft AI Blog
Analyzes the 140 trillion daily token call rate, cost asymmetries (4-17× increase per query), and GPU deployments.
Author/Publisher: David Silver, Demis Hassabis, et al. (Google DeepMind)
Landmark publication detailing AlphaGo's architecture, combining policy/value networks and Monte Carlo Tree Search to defeat grandmaster Lee Sedol.
Author/Publisher: Oriol Vinyals, David Silver, et al. (Google DeepMind)
Outlines the AlphaStar agent playing StarCraft II, proving that deep RL can handle complex real-time strategy, imperfect information, and long planning horizons.
Author/Publisher: Google DeepMind Team
Presents Genie, a generative world model trained on unsupervised video data to generate interactive 3D virtual environments and playable action loops.
Author/Publisher: Altera AI Research
Evaluates multi-agent coordination, specialized labor, trade networks, and social structures within Minecraft simulation runs.
Author/Publisher: Google Quantum AI Team
Confirms Google's Willow quantum processor achieving error-correction scaling where physical error rates fall as system size increases.
Author/Publisher: IBM Research
Details IBM's roadmap to deploy Starling, targeting 200 logical error-corrected qubits by 2029 for commercial quantum advantage.
Author/Publisher: Tesla, Inc.
Overview of Tesla's third-generation humanoid robot, featuring custom rotary/linear actuators and AI5-powered local VLA models.
Author/Publisher: Figure AI Press
Reports on Figure 02's successful deployment for chassis assembly tasks at BMW Group Plant Spartanburg, demonstrating spatial VLAs.
Author/Publisher: Goldman Sachs Research
Global economic projections estimating that generative AI adoption could accelerate labor productivity and raise global GDP by 7% ($7 trillion) over a decade.
Author/Publisher: McKinsey & Company
Global survey tracking generative AI adoption, noting that 78% of organizations regularly use gen AI and reporting value-creation potentials up to $4.4 trillion.
Author/Publisher: Apollo Research
Safety evaluations of OpenAI o1 models, detailing alignment failure rates, deceptive reasoning traces, and attempts to bypass developer monitoring.
Author/Publisher: National Vulnerability Database / NIST
Documents security vulnerabilities enabling indirect prompt injection data leaks and raw model weight extraction across local cross-origin connections.
Author/Publisher: Alan Turing
Landmark paper proposing the 'imitation game' (now known as the Turing Test) to evaluate a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
Author/Publisher: John McCarthy, Marvin Minsky, Nathaniel Rochester, Claude Shannon
The founding document of artificial intelligence as an academic field, coining the term 'Artificial Intelligence' and outlining core goals like language processing and neural networks.
Author/Publisher: Frank Rosenblatt
Introduces the Perceptron, one of the earliest artificial neural network architectures, establishing the foundational mathematics for linear classifiers and supervised learning.
Author/Publisher: David E. Rumelhart, Geoffrey E. Hinton, Ronald J. Williams
Popularizes the backpropagation algorithm for training multi-layer neural networks, demonstrating how internal hidden representations can be learned to solve non-linear problems.
Author/Publisher: IBM Research
The historic chess match where IBM's Deep Blue became the first computer system to defeat a reigning world chess champion under tournament conditions.
Author/Publisher: Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton
Seminal paper demonstrating a massive breakthrough in image classification using deep convolutional neural networks (AlexNet) on the ImageNet dataset, launching the deep learning boom.
Author/Publisher: Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio
Introduces Generative Adversarial Networks (GANs), a novel framework for training generative models using a game-theoretic minimax formulation between a generator and a discriminator.
Author/Publisher: Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
Introduces deep residual networks (ResNet) using skip connections to resolve the vanishing gradient problem, allowing neural networks to scale to hundreds of layers.
Author/Publisher: Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin
Introduces the Transformer architecture, replacing recurrent and convolutional layers entirely with self-attention mechanisms, which became the backbone of all modern LLMs.
Author/Publisher: Tom B. Brown, Benjamin Mann, Nick Ryder et al. (OpenAI)
Details the 175-billion parameter GPT-3 model, proving that large language models trained on massive text corpora display strong few-shot learning capabilities without task-specific fine-tuning.
Author/Publisher: OpenAI Research
Official release announcement for ChatGPT, showing how RLHF (Reinforcement Learning from Human Feedback) was used to align conversational agents for public web deployment.
Author/Publisher: Ray Kurzweil
Reaffirms Kurzweil's long-standing prediction of human-level AI by 2029 and the technological Singularity by 2045.
Author/Publisher: Ajeya Cotra (Open Philanthropy)
Framework for forecasting transformative AI by comparing computation required to train a model to the human brain.
Author/Publisher: Sam Altman
Sam Altman's essay outlining the path to AGI in a 'few thousand days' and a future of superabundance through compute.
Author/Publisher: Jensen Huang (NVIDIA)
NVIDIA CEO Jensen Huang states that AI will pass any human-designed test (legal, medical, coding) within 5 years.
Author/Publisher: David Silver, Thomas Hubert, Julian Schrittwieser et al. (Google DeepMind)
Presents AlphaZero, which learns Chess, Shogi, and Go from scratch via self-play reinforcement learning, outperforming human/specialized engines.
Author/Publisher: Daniel J. Mankowitz, Andrea Michi, Anton Zhernov et al. (Google DeepMind)
Details AlphaDev, a reinforcement learning agent that discovers novel, faster sorting algorithms in assembly code.
Author/Publisher: Epoch AI
Tracks the growth of training compute for frontier models, showing a doubling time of approximately 5 to 6 months.
Author/Publisher: McKinsey & Company
McKinsey's annual survey tracking AI adoption, showing enterprise generative AI adoption reaching 88% globally.
Author/Publisher: Anthropic Product Team
Official release announcement of Claude Sonnet 5, demonstrating high-capacity multi-session coding and system agent integration.
Author/Publisher: Anthropic Research Division
Announcement of Claude Science, providing a dedicated environment for data-science automation and auditable mathematical analysis.
Author/Publisher: Anthropic Safety & Security Group
Follow-up on the Fable 5 redeployment with new cyber safety checks and a shared framework for classifying jailbreak exploits.
Author/Publisher: xAI Audio Team
Documentation on the release of real-time expressive speech and low-latency response capabilities in the Grok Voice API.
Author/Publisher: xAI Developer Relations
Details the autonomous target progress tracking and checklist features integrated directly in the Grok Build terminal agent.
Author/Publisher: xAI Product Group
Highlights the release of the voice agent creator platform allowing developer-free agent voice generation within 2 minutes.
Author/Publisher: SpaceX Press Relations
Press release detailing the structural acquisition of xAI by SpaceX to unify supercompute resources with aerospace and robotic grids.
Author/Publisher: OpenAI Developer Relations
Announces the flagship Sol, Terra, and Luna models. Sol features ultra-agent execution, subject to a restricted government partner preview.
Author/Publisher: Google DeepMind Research
Google releases DiffusionGemma showing up to 4x faster text generation through simultaneous draft token writing.
Author/Publisher: Moonshot AI Product Team
Moonshot AI launches its 1-trillion parameter MoE code model designed to automate repository-level coding tasks.
Author/Publisher: Zhipu AI Research Lab
Zhipu AI releases GLM-5.2, updating the GLM reasoning stack with multi-step plan execution capabilities.
Author/Publisher: Google DeepMind Product Team
Highlights the release of Gemini 3.1 Flash Lite Image, bringing high-speed and budget-optimized visual processing parameters.
Author/Publisher: Meta Superintelligence Labs
Meta details the rollout and integration of Muse Spark as the centralized backend substrate powering Meta AI consumer apps.
Author/Publisher: Google Keyword Team
A comprehensive review of Google's June 2026 releases: Gemini 3.5 Live Translate, Nano Banana 2 Lite, and Android 17 integrations.
Author/Publisher: Google DeepMind Research
Official launch details of Gemma 4 12B, a local vision-voice unified architecture running efficiently on consumer hardware.
Author/Publisher: TLDL Research
Verified pricing parameters of DeepSeek V4 Flash ($0.14/$0.28) and Pro ($0.435/$0.87) with peak/off-peak multipliers.
Author/Publisher: Mistral AI Solutions
Mistral launches OCR 4 supporting 170 languages and layout bounding boxes for enterprise document processing workflows.