Skip to main content
Back to News
news/AI Infrastructure

Meta's strategy chief jumps to inference startup Fireworks

Meta strategy chief Dave Wehner leaves after 14 years to become CFO of Fireworks, the $17.5 billion inference startup scaling open-model serving.

Stefan Trbojevic

Stefan Trbojevic

8 November 20263 min read
LinkedIn
Empty dark data-center room lit by glowing teal wall panels

The takeaway

The inference layer is maturing into enterprise infrastructure: talent, capital and pricing discipline are moving there, and builders should expect vendor contracts to get more formal.

Why it matters for builders

Serving-layer vendors are becoming enterprise suppliers. Expect procurement, SLAs and security reviews, margin-driven pricing changes, and higher stakes on credential hygiene in evaluation and CI environments.

Meta's strategy chief jumps to inference startup Fireworks

Meta's chief strategy officer is leaving the company after 14 years to run the finances of a much smaller company that sells the pipes AI models run through. Dave Wehner, who was Meta's chief financial officer from 2014 to 2022, announced on Friday that he is joining Fireworks as CFO, the company confirmed to CNBC.

What happened

Wehner joined Facebook in 2012 as a vice president of corporate finance and moved into the CFO seat two years later. Fireworks chief executive Lin Qiao said Wehner will be responsible for scaling the company's finances and operations, while Meta chief operating officer Javier Olivan credited him with bringing "disciplined thinking" to some of Meta's hardest company-level decisions.

Fireworks was founded in 2022 by former Meta and Google engineers. It runs an inference and model-training platform that lets companies fine-tune open models on their own data and serve them in production, without building GPU infrastructure from scratch. The company raised $1.5 billion in July at a $17.5 billion valuation, with Nvidia and Sequoia Capital among its backers. CNBC, citing Business Insider, reports that Fireworks is in early talks to raise again at a valuation of $30 billion to $40 billion.

Luminous filaments passing through a glowing glass block on a dark surface representing token traffic through an inference layer

Why the inference layer is hiring finance chiefs

Inference is where token volume turns into revenue, and the volume is enormous. At its July round the company said it had passed $1 billion in annualised revenue and was processing more than 40 trillion tokens a day, with over 95% of that traffic coming from models customised on customer data rather than off-the-shelf open weights.

That mix changes the job. A company selling fine-tuning and dedicated GPU capacity at that scale needs capital planning, contract discipline and pricing strategy. Wehner is a public-company-grade finance operator moving into a private infrastructure vendor that is heading toward much larger commitments.

The move is not frictionless. In October, Fireworks disclosed a security incident in which an unauthorised third party used internal credentials to reach environment variables connected to evaluation jobs in a limited number of customer accounts. The company said its investigation had found no evidence of unauthorised access to customer models, datasets, training data or inference traffic.

What this means for builders

For teams running open models in production, the signal is that the serving layer is consolidating into serious vendors rather than hobby-scale endpoints.

  • Expect harder enterprise sales motions: procurement, SLAs and security questionnaires, not just an API key.
  • Watch pricing. Vendors buy GPU capacity and resell tokens, so price moves are a leading indicator of their margins.
  • Treat credentials as production risk. The October incident involved evaluation-job environment variables, the kind scattered across sandboxes and CI pipelines.

The infrastructure picture is shifting around the model layer, including moves like NVIDIA open-sourcing a GPU Kubernetes control plane. Who controls serving capacity is becoming as important as who trains the best model.

Share𝕏

The Automation Brief

Read 5 AI stories instead of 50.

The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.

No noise. Unsubscribe anytime.

Editorial notes

Reported by

Stefan Trbojevic

Edited by

n8n Lab Editorial

Published

8 November 2026

Updated

8 November 2026

AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.

n8n Lab is an independent service provider. We are not affiliated with, endorsed by, or sponsored by n8n GmbH. “n8n” is a trademark of n8n GmbH and is used here only to describe the platform-specific implementation and automation services we provide.