Turn a rough research question into a fully sourced Deep Research report

Use Deep Research for this. Here's the scope — don't loosen it once you're running.

RESEARCH QUESTION
Is on-device AI inference becoming cost-competitive with cloud inference for mid-size SaaS companies in 2026?

MUST COVER
hardware cost trends, latency tradeoffs, at least two vendor case studies, and total cost of ownership over 2 years

SOURCE PREFERENCES
Prioritize: vendor technical documentation, published benchmarks, and analyst reports from named firms
Treat these only as context, not primary evidence — mention them only if you're also verifying the claim elsewhere: marketing blog posts and unverified Reddit/forum claims

OUT OF SCOPE
consumer-device on-device AI (phones/laptops) — this is about server-side SaaS inference only — don't wander into these even if a source brings them up.

BEFORE YOU FINALIZE
- Open primary sources where possible (original studies, filings, official documentation), not just aggregator summaries of them.
- Where sources disagree, say so and name both positions instead of picking the one that sounds more authoritative.
- Note the publication date of anything time-sensitive — don't present an older figure as current without flagging it.

OUTPUT
Structure the report as: executive summary, then one section per must-cover angle, then a recommendation, then sources
End with a sources list, each one labeled by how you used it (primary evidence / context / disputed claim).