[subscription] Applied Computing, a three-year-old London startup developing foundation models for energy operations, raised $20 million round led by KBR with Databricks Ventures.. Applied Computing. Chinese AI Startup Takes on Anthropic and OpenAI with New Model

Decentralized GPU cloud infrastructure for AI and gaming. Enterprise-GPU-first. Competitor to Feya in decentralized compute.

Decentralized AI compute network. New York NY + Los Angeles CA. Splits frontier models across participant-owned GPUs over public internet. FEYA token = unit of account: 1 FEYA = 1M completion tokens, flat rate across all hardware. Two-layer architecture: off-chain commerce (instant, zero fees) + on-chain token (post-TGE). Asymmetric bridge: deposits instant, withdrawals gated. Three product surfaces: Chat (consumer), API (developers), Contribute (GPU providers). Live: 3-node network NYC + LA as of April 2026. Raising seed: 1-5% equity, ~$1-5M. Token: 100B FEYA genesis supply at TGE. Corporate structure: Delaware C-Corp (Feya AI) + Cayman Foundation (token issuer). Competitors: Akash, Render, IO.NET, Bittensor, Aethir — but closest is Gonka. Distinction: consumer-hardware-first + real AI product on top. Confidential document: Economic Architecture v1, April 2026. Osparna active deal — introduced by Audrey McKnight (Matt McKnight daughter) May 2 2026. [2026-05-25 per fact 427] Founded by Audrey McKnight (persons.id=637), Matt McKnight's daughter. Pitch sent to troy@osparna.com May 2 2026 with Matt asking for Troy's take. Re-tenanted from 1 (public) to 3 (Osparna) per fact 426 role-reading: pitch addressed to Troy's Osparna-investor role. Deal-substrate currently in evaluation. Seed round $1-5M. TECHNICAL ARCHITECTURE DETAIL (from Feya architecture.html + Economic Architecture v1 PDF, read directly 2026-07-09): Three inference paths on one OpenAI-compatible API -- Standalone (live, one node/one 7B-13B model), Cluster (built not yet demonstrated end-to-end, pipeline-parallel across workers via TLS-PSK RPC, 5% lead bonus), Backbone (Phase 2, protocol-owned H100/B200 datacenter hardware). Reward formula: work_units = completion_tokens x quant_multiplier x speed_bonus; provider_reward = work_units x base_rate x (1 - operational_emission 8%). Coordinator on AWS EC2 us-east-2; nodes run under launchd with SSH reverse tunnels -- Audrey's own Mac (M2, 24GB, Metal, NYC) is one of the three live production nodes, alongside a CFO laptop (NYC, RTX 4060) and CEO laptop (LA, RTX 3070). ECONOMIC ARCHITECTURE (confidential doc, "Accredited investors," April 2026): flat compute rate 1 FEYA = 1,000,000 completion tokens, immutable, hardware/geography/demand-independent. Two-layer settlement: off-chain (elastic, instant, zero-fee commerce) + on-chain (100B genesis ceiling, active post-TGE). Asymmetric bridge -- deposits instant, withdrawals gated/linear-unlocked -- by design, to buy the protocol time against panic exits. Spend multiplier k (launch 2.0x) charges above-free users; providers always earn (1-f) flat regardless of k or user level -- provider income deliberately isolated from every governance lever, since supply-side (node operator) churn is named as the network's most fragile property. Genesis distribution: 100B FEYA -- Ecosystem 25%, Treasury 20%, Founders (3) 20% (4-yr vest/12-mo cliff), Future Team 7%, Reserve 7%, Private Investors 6%, Protocol-Owned Liquidity 5%, Launch Liquidity 5%, Advisors 3%, Public Sale 2%. Corporate split: Delaware C-Corp (Feya AI, operating company + its own independent AI product revenue) and Cayman Foundation (token issuer/steward, holds treasury, pays C-Corp via intercompany service agreement) -- equity and token deliberately non-overlapping instruments, modeled on the Solana Labs/Solana Foundation structure. Fundraising stack: Seed (equity, 2-5%) -> Private Round I/II (token, Reg D 506(c)) -> optional Series A (equity) -> Public Sale at TGE (Reg S). METHOD FAILURE, LOGGED HONESTLY: I minted a duplicate company row for this (id 8050, since deleted) before finding this one -- checked companies by name "Faye" (wrong spelling, before the correction) and t0_corpus/agent_docs, but never checked individuals.current_company_id on Audrey's own record (637), which pointed here directly. A live recurrence of [[principle-retrieval-failure-disqualifies-osparna]] (id=94), minutes after diagnosing it and after being told directly to check tenant 3.

[subscription] Dave Brown, longtime Amazon Web Services executive who ran AWS' compute and machine learning services and the Trainium AI chip business, is departing the cloud provider. Dave Treadwell from Amazon's commerce group will replace him. The move comes as AWS accelerates revenue growth serving large AI customers like Anthropic and OpenAI.. Amazon Web Services (AWS). Fwd: TSMC Plans Another $100 Billion U.S. Investment as AI Demand Surges

AI safety and research company; develops the Claude family of AI models. Project Glasswing is an Anthropic program giving a vetted set of organizations limited access to its Claude Mythos model for cybersecurity testing; expanded to 150 organizations across more than 15 countries and industries including power, water, and healthcare (June 2026). Anthropic has been criticized for restricting Mythos to a narrow group and has said it intends to release the model publicly once safeguards against cyberattack misuse mature. Filed confidentially for an IPO (June 2026). Public facts per The Information, 2026-06-03.

[subscription] Ayar Labs, which makes equipment for ultrafast connections between AI chips, is raising capital at a sharply higher valuation than previously.. Ayar Labs. Fwd: FuriosaAI, Nuvacore, D-Matrix Join Chip Funding Spree

[subscription] AMD announced a partnership with Cerebras, a rival AI server chip developer, to connect AMD's server racks to Cerebras wafers to run both companies' chips simultaneously. This is an example of disaggregated inference where AMD's GPUs handle prefill (compute-intensive) and Cerebras' chip handles decode. The combined product will launch later this year. Cerebras shares jumped 7% on the announcement.. Cerebras. Fwd: Stripe in Talks to Buy Startup OpenRouter

Wafer-Scale Engine AI chip maker. IPO May 14 2026 (CBRS). $510M 2025 revenue +76% YoY. $238M net income. OpenAI $20B+ customer.

[subscription] Etched, an AI chip developer, is negotiating two funding rounds: Sequoia-led at $10B valuation and Jane Street-led at $20B valuation.. Etched. Fwd: AI's true cost