Startup profile for Zyphra: latest funding, latest known valuation, total disclosed funding, investors, status, sources, and why the company is interesting. Part of the Venture Capital Tracker startup directory.

Startup profile · funding coverage

Zyphra funding, valuation and investors

Open-model lab and AMD-centered AI cloud developer behind Zamba, ZAYA and a diffusion-converted 8B model.

Verified company research · funding, investors and latest developments

Keep track of Zyphra

Save this profile to your VCT watchlist for a quick return.

View watchlist

Funding, valuation & investors

Answer-first snapshot

Latest funding

Reported Series A

$110M (Forbes citing PitchBook; Dealroom also records $110M) · June 2025

Latest known valuation

$1B, market-data reported

Reported Series A · June 2025

Total disclosed equity funding

$110M

Excludes debt, grants, acquisitions, secondaries, and IPO proceeds.

Current status

private

series a · San Francisco Bay Area, United States

Investors in latest funding

Other: Future Ventures , Bison Ventures , Metaplanet / Jaan Tallinn , AMD Ventures

Sources: latest funding · tollbit.forbes.com Last verified: 2026-11-13

Overview

Zyphra develops open-weight hybrid state-space/Transformer models, foundation models trained on AMD accelerators and AMD-focused AI infrastructure. Its ZAYA1-8B-Diffusion-Preview was published in May 2026 and uses block diffusion to generate 16 token drafts simultaneously. Zyphra claims up to 4.6x faster generation with its lossless sampler and 7.7x with a faster sampler that can affect quality.

Forbes, citing PitchBook, reported $110M raised at a $1B valuation in 2025. Forbes reported Zyphra was raising a $500M Series B at at least $5B in May 2026; the company declined comment and a completed closing remains unconfirmed.

Why Zyphra is interesting

Direct NCM-adjacent technical competitor: Zyphra's May 2026 ZAYA1-8B-Diffusion-Preview generates blocks of 16 tokens in parallel, claiming 4.6x or 7.7x acceleration with different sampler quality tradeoffs. It is not Stanford's NCM.

Product & use cases

Zamba hybrid SSM models, ZAYA MoE reasoning models, a research-stage discrete diffusion-converted language model and Zyphra Cloud on AMD GPUs.

  • Parallel block-diffusion language generation
  • Efficient open-model training and inference on AMD accelerators
  • Open-weight model deployment and custom AI infrastructure

Key facts

  • May 2026 ZAYA1-8B-Diffusion-Preview: 16-token blocks, up to 4.6x/7.7x company-reported decoding speedups
  • 2025 $110M at $1B valuation reported by Forbes/PitchBook
  • $500M Series B reported being raised May 2026, not confirmed closed

Funding history (newest first)

Reported Series A

2025-06 (reported; precise close date unverified) $110M (Forbes citing PitchBook; Dealroom also records $110M) Valuation: $1B, market-data reported

Source: Read source · tollbit.forbes.com

Competitive landscape

Edge: End-to-end control of an alternative hardware stack, open-weight models and diffusion decoding research; performance is company-benchmarked against a specific baseline.

Unique crossover between diffusion LLM research and infrastructure. Its comparative 4.6x and 7.7x numbers depend on the sampling method and hardware, not a general guarantee. $500M potential 2026 Series B is prospective, not confirmed proceeds.

Market / IPO context

Editorial / static context — not a live quote.

FAQs about Zyphra

Practical answers founders, operators, and investors typically search for.

Yes. Its ZAYA1-8B-Diffusion-Preview uses blocks of 16 tokens with different sampling trade-offs; this is a different approach from Stanford's NCM.
Forbes reported fundraising discussions in May 2026, not a confirmed completion. VCT does not include that proposed $500M as invested capital.
Forbes, citing PitchBook, reported a $110M financing at a $1B valuation in 2025; this was not verified through a company announcement.

By Venture Capital Tracker

Last updated:

Editorial note: AI tools assisted with research, structure, or drafting. Venture Capital Tracker retains human editorial responsibility for factual accuracy, relevance, and source quality before publication.