WEKA's new inference product marketer is asked to "quantify WEKA's economic value across GPU utilization, time-to-first-token, throughput, concurrency, tokens per GPU, and cost per token". The site already promises that number, but only after a form and a 30-minute call ("we'll send you an estimate of what your cost per token will be"). The two calculators that are self-serve answer older questions: one prices WEKApod storage for a SuperPOD from a GPU type and count, the other uses cloud list prices "as of May 30, 2023". Neither mentions tokens.
First action
Write the formula from the OCI benchmark and test it on three known customer setups.
How to measure it
Inference cost-analysis meetings
- Now
- –
- Target
- 2x current
- By
- 90 days
Steps
Agree the model
Inputs, outputs and the benchmark ratios behind them.
Senior Product Marketing Manager, AI Inference 1 week
Check the method
Engineering signs off the assumptions and ranges.
Principal Product Manager, Augmented Memory Grid 3 days
Design
One screen: inputs left, cost per million tokens right.
Web team 3 days
Build
On the AI Inference page, with results shown before any form.
Web team 2 weeks
Pass inputs to the form
Prefill the 30-minute analysis request.
Marketing Operations 2 days
Retire the old pair
Point the SuperPOD and 2023 cloud calculators to the new one.
Web team 1 day
Track
Runs, inputs chosen and meetings booked.
Marketing Operations weekly
Why it works
Build an inference economics calculator on /solutions/ai-inference-acceleration: inputs for model size, context length, concurrent users, GPU type and count, and current KV-cache setup; outputs tokens per GPU, time-to-first-token and cost per million tokens with and without Augmented Memory Grid, using the published OCI and CoreWeave benchmark ratios with the method shown. Show the result immediately, then offer the 30-minute analysis to check it against the visitor's own cluster, with their inputs passed to the form.
Inference buyers already think in cost per token, and a number they can play with earns the meeting instead of being the price of it. Every run also tells the inference marketer which models, context lengths and GPU fleets prospects care about.
Ready to use
ChecklistCalculator inputs and outputs
[ ] Model size (8B / 70B / 405B / custom) [ ] Context length and prompt reuse [ ] Concurrent users [ ] GPU type and count [ ] Current KV-cache tier (HBM only / DRAM / NVMe) [ ] Output: tokens per GPU, time-to-first-token, cost per million tokens [ ] Method note citing the OCI and CoreWeave benchmarks [ ] "Check it with an engineer" button prefilled with the inputs
Risks and how to handle them
Numbers read as a promise
Show ranges, the benchmark source and a "your results will vary" note.
Competitors copy the inputs
The value is the method and benchmark data, which only WEKA has.
Evidence
- HiringSenior Product Marketing Manager, AI Inference (posted September 28, 2026): "Define and quantify WEKA's economic value across GPU utilization, time-to-first-token, throughput, concurrency, tokens per GPU, and cost per token. Build the economic case customers need to justify investment." The role also recruits "design partners and early adopters", owns the proposition for AI clouds that should make WEKA "the storage standard for the category", and prefers to "show specific numbers and named customers than make broad claims".Source
- Your websiteAI Inference Cost Analysis page: "In a 30-minute call, we'll identify where your inference costs are highest"; "Once we get a snapshot of your infrastructure, we'll send you an estimate of what your cost per token will be on NeuralMesh"; proof points "Stability AI reduce their storage costs by 95% on a per TB basis" and CoreWeave benchmarks with "up to 4.2x" token throughput per GPU.Source
- Your websiteThe two calculators on the site: the TCO Calculator ("See how much you can save by using WEKApod as the storage for your SuperPOD deployment") takes only a GPU type (H100/200, B200, GB200) and a GPU count and shows two totals, e.g. "$73,203,001 WEKA" against "$108,203,001 Other", and has no meta description; the Cloud Savings Calculator says "Results are based on publicly available list prices as of May 30, 2023" and asks for a form before showing savings. Neither mentions tokens, time-to-first-token or cost per token.Source
- News coverageWEKA newsroom: NeuralMesh 6 (July 21, 2026) cites OCI benchmarks of "10x higher token throughput, 10x more concurrent users served, and 7x more tokens per GPU"; WEKA raised a $140 million Series E at a $1.6 billion valuation.Source