Fireworks AI

fireworks.ai ↗added 2026-06-19

★ Nvidia participated● Hiring

TL;DR中文

今年infra层最猛的故事之一：7个月估值从$4B → $17.5B，收入一年5x到$1B+ run rate。押注的论点是"企业最终会用自己数据训的specialized模型，而不是一直租通用大模型"。

Worth-Joining Index · Scorecard

v2.1 ALPHA · 2026-08

Career Score

84/100

Coverage

89%

Verified

40%

Stage cohort

Late

Vibe Check

🎪 community voting coming soon

Momentum

16/16

Efficiency

14/14

Backing

5/5

Market

8.8/11

Product

8.8/11

Upside

8.5/16

See all rankings →Methodology →Company or employee? Correct the data →

Latest valuation

$17.5B

Series D · 2026-07

Annualized revenue

Over $1B run rate

✓ Publicly verifiable ↗as of 2026-07

Team size

~150 people, planning to reach 600 by end of 2026

AI InfraInferenceFounded · 2022Stage · Series D ($17.5B)HQ · Redwood CitySize · ~150 people, planning to reach 600 by end of 2026

Traction · Key numbers

Annualized revenueOver $1B run rate✓ Publicly verifiable ↗as of 2026-07

Revenue growth5x YoY✓ Publicly verifiable ↗as of 2026-07

ThroughputProcesses 40 trillion (40T) tokens per day — vs ~27T/day for Google's developer API and ~22T/day for OpenAIA four-year-old startup out-throughputs two giants' developer businesses; the Google/OpenAI comparison figures come from the CEO's own interviews✓ Publicly verifiable ↗as of 2026-07

Specialized share95%+ of tokens come from models customers have specialized with proprietary data, not generic model callsThe single most convincing data point for the whole thesis◆ Company-claimedas of 2026-07

Funding

Rounds

Valuation pace

$4B → $17.5B

$1.505B

Series D · 2026-07 · valuation $17.5B · led by Atreides, Index, TCV

With: Nvidia, Lightspeed, Bessemer, Menlo等

$250M

Series C · 2025-10 · valuation $4B

Nine months before the Series D

▸Nvidia participated

▸Microsoft partnership signed in March; compute sourced from 20+ suppliers

Founders

Lin Qiao CEO

Founding team came out of Meta's PyTorch team — she previously ran PyTorch at Meta; in inference optimization the technical pedigree is about as good as it gets

Reality · Working here

H1B / PERMUnknown

Remote policyUnknown

Comp modelUnknown

GlassdoorUnknown — work here? Help us fill this in →

Equity liquidityUnknown

Hiring statusThis is its most aggressive hiring year yet — core technical roles are performance optimization (GPU kernels, distributed systems, LLM/VLM serving); on the GTM side, building a sales org from zero (hiring everything from BDR to strategic finance)✓ Publicly verifiable ↗as of 2026-07

·Founding team from Meta PyTorch + Google Vertex AI — strong engineering culture, self-described as no bureaucracy, just results

·150 people carrying $1B revenue means extreme per-head efficiency — the flip side is few people and lots of work; don't expect work-life balance

·Poached former Salesforce exec George Hu as President in April — growth used to be mostly customer self-serve, now they're building a real sales team, a sign they've entered the scale phase

Signals & Risks

CompetitorsTogether AI ($8.3B valuation, ~$1.15B bookings)Baseten ($5B; differentiated by in-VPC self-host/hybrid deployment, an edge in compliance-heavy industries)Replicate and assorted neocloudsPressure from the hyperscalers — Microsoft has its own Foundry (a partner for now), and Nvidia is moving up the stack

Gross marginGross margin is only ~50% (GPU costs sit in COGS), far from a software company's 70%+; management's own target is just 60%

CommoditizationToken prices keep falling, so they must out-optimize the pace of price cuts; the risk is commoditization keeps margins permanently capped

Valuation~17.5x ARR isn't cheap (though the multiple is healthier than Harvey's)

CompetitionAn extremely crowded field; the endgame may come down to capital efficiency and who locks in the big customers first

Editorial ✎

Good fit for

· Infra/systems engineers — you get to work on the largest production inference workloads in the industry

· Commercial talent who want to help build GTM during hypergrowth (George Hu just arrived as President, so early commercial roles have big growth headroom)

· People who value transferable skills — even if the company doesn't make it, anyone who has run a 40T tokens/day system is in demand everywhere

· People who want in before the org balloons — this is its most aggressive hiring year, so the timing is good

Not for

· People counting on work-life balance

Sponsorship

✓ Publicly verifiable · government data

H-1B recent sponsorship activity Latest

DOL LCA Disclosure Data · FY2024 – FY2026 Q2 ↗

No LCA records in FY2024 – FY2026 Q2 — no public H-1B sponsorship activity recently. Confirm directly in interviews, or ask insiders in the group.

H-1B approval history

USCIS H-1B Employer Data Hub · FY2021–2023 ↗

No approvals in FY2021–2023. Most companies on the list were founded after 2022, so missing early records is normal — it doesn't mean they don't sponsor today. Confirm in interviews, or ask insiders in the group.

PERM green cards

DOL PERM Disclosure Data · FY2025 – FY2026 Q2（至2026年3月） ↗

No records for this company in the imported PERM data. Green-card processes usually start once a company matures.

Want to check another company?Sponsorship lookup →

🔒

Editorial deep dive · Members only

The full research writeup (background, key numbers, Bull & Bear, joining advice, and an interview question checklist) is reserved for community members. Members: enter the access passphrase from the group.

Not a member yet?Request to join →

Want to talk about this company?

Community members may work here or have interviewed — corrections and additions welcome too

Join the discussion