Tags · #inference

Nvidia, AMD, and CoreWeave all backed Tensormesh — KV-cache reuse becomes an inference primitive

Read this because Three rivals — Nvidia, AMD, CoreWeave — co-investing is the tell: KV-cache reuse (don't recompute what you already computed) is being treated as a foundational, neutral inference-stack layer. The economics of the inference era, in one round.

Tensormesh raised $20M from Nvidia, AMD, and CoreWeave and shipped Tensormesh Inference — productized KV-cache reuse claiming up to 10x lower latency/GPU cost.

OpenRouter raises $113M to own the switchboard between every AI model

Read this because The bet isn't a model — it's the layer above all of them. OpenRouter sells optionality: route to whatever model is best/cheapest, lock into none. The tell is the cap table — Nvidia plus Snowflake, Databricks, MongoDB, ServiceNow all want a seat at the routing layer.

OpenRouter raised a $113M CapitalG-led Series B at ~$1.3B (May 26) — routing 100T tokens/month across 400+ models for 8M+ users.

Anthropic in talks to rent Microsoft's Maia 200 AI chips — compute-crunch hedge

Read this because Silicon diversification, not a chip win. Anthropic already runs on Nvidia, Google TPUs, and AWS Trainium — adding Maia 200 makes it the first lab spanning all four silicon families. Optionality is the moat when compute is the bottleneck.

Anthropic is in talks to run Claude inference on Microsoft's Maia 200 chips via Azure (no deal signed, per CNBC May 21) — a hedge away from Nvidia + TPUs.

Cerebras drops 10% on Day 2 after 68% IPO pop — OpenAI-concentration risk lands

Read this because First public verdict on whether the market believes Nvidia has a credible challenger — or whether Cerebras' OpenAI/AWS revenue concentration was priced for a fall.

CBRS closed Friday at $279.72, -10% from Thursday's $311.07 debut. Still +51% over the $185 IPO price; market cap ~$60B on 215M shares outstanding.

Cerebras CBRS opens at $350, closes +68% in 2026's biggest IPO debut

Read this because An IPO pop isn't a moat. The real test is whether +68% holds after the 6-month lockup — and whether anyone actually swaps NVDA inference for CBRS.

Cerebras priced at $185, opened at $350 (+89%), closed $311.07 (+68%). Implied market cap roughly $70B fully diluted. $5.55B raised, 20× oversubscribed.

Cerebras hikes IPO range to $125–$135 as orders hit ~$10B against $3.5B on offer

Cerebras hikes IPO range to $125–$135 as orders hit ~$10B vs $3.5B on offer. Implied valuation comfortably above $26.6B; CBRS prices May 13.

DeepSeek V4, Kimi K2.6, GLM-5.1, MiniMax M2.7: frontier-tier coding at 5–25× lower cost

Four Chinese labs shipped frontier coding models in 12 days. GLM-5.1 MIT license ties Kimi K2.6 on SWE-Bench Pro at 58.4%. Cost gap is 5–25× vs Western APIs.

Cloudflare Infire — disaggregated LLM inference beats vLLM by 20%, Unweight cuts model size 22%

Cloudflare Infire (Rust) uses disaggregated prefill/decode to beat vLLM 0.10 by 20% on H100s. Unweight achieves 15–22% lossless model weight compression.

vLLM v0.20.0 ships DeepSeek V4 + PyTorch 2.11 + FlashAttention 4

vLLM v0.20.0: 752 commits, 320 contributors. CUDA 13, PyTorch 2.11, Transformers v5, Python 3.14, FlashAttention 4 default, 2-bit KV cache.