How much does an AI consultant cost? The AI spend ladder
AI consultant cost in the US has no national survey: Upwork's published bands, $35 to $200 an hour, cover worldwide contracts. Buy the lowest rung you can use.
Angus McDonald · 14 Aug 2026 · 11 min read
written for United States ● also for Australia, United Kingdom, Canada, New Zealand

AI consultant cost in the US runs $35 to $200 an hour on Upwork's published marketplace rates for AI engineers and machine learning engineers, which Upwork's own footnote says cover historical contracts worldwide rather than US-only work. Upwork's project bands run from about $500 for a basic chatbot to $20,000 for a custom model, in US dollars. No authoritative US rate survey for AI consulting exists.
What AI consultants charge per hour, per day and per project in 2026
What it costs to hire an AI consultant in the United States has no single authoritative source. No government rate table covers AI consulting and no industry body publishes one, so every band below carries its provenance in the row.
| What you are buying | Band in US dollars | Where the number comes from |
|---|---|---|
| Solo contractor, AI engineer, hourly | $35 to $60/hr, median $50 | Upwork marketplace rates, worldwide contracts in USD |
| Solo contractor, ML engineer, hourly | $50 to $200/hr, median $100: beginner $50-$80, intermediate $80-$120, advanced $120-$200+ | Upwork marketplace rates, worldwide contracts in USD |
| An employed equivalent, for comparison | median $101,190 a year; bottom 10% under $59,720, top 10% over $174,140. The $48.65/hr equivalent is arithmetic at 2,080 hours, not a published figure | BLS, management analysts, May 2024 reference period, about 27 months old |
| Fixed-price project | chatbot on third-party software $500 to $1,000; custom ML model $8,000 to $20,000; process automation from about $6,000; custom development $5,000 to hundreds of thousands | Upwork project cost guidance |
AI consultant pricing splits into three tiers, and only the first is sourceable. A solo contractor's ceiling is about $200 an hour, published. A boutique firm's floor sits above it and is published nowhere verifiable: the hourly ranges quoted for that tier trace back to vendor marketing pages citing each other. Large firms publish nothing at all for AI engagements. The tiers differ in risk allocation rather than quality. Hours leave the scope risk with you; a fixed fee moves it to the vendor.
No verified US day rate for AI consulting exists either. American buyers are quoted hourly or per project, and day rates are a UK and Australian convention. Eight hours at the marketplace band is roughly $400 to $1,600 a day, which is arithmetic from the hourly source, not a survey. The nearest thing to public US rate data is GSA CALC, which publishes fully burdened not-to-exceed ceiling rates awarded under GSA Multiple Award Schedule contracts. It covers federal contract ceilings rather than the small-business market.
The Azgard AI Spend Ladder: five rungs and five units of purchase
The Azgard AI Spend Ladder sorts AI consulting spend by the unit being bought, not by the size of the vendor selling it. The unit decides who pays when discovery finds a different problem than the one in the brief.
| Rung | Unit bought | What you own at the end | Who absorbs the cost if scope is wrong | Price band (US dollars) |
|---|---|---|---|---|
| 0, below the ladder | Ten minutes a day of your attention, for a week | Enough vocabulary to judge a proposal | Nobody: there is no scope | $0, the Department of Labor's free AI literacy course |
| 1 | An hour | Notes, and whatever was said in the hour | You. Every hour spent learning your business is billed to you | $35 to $200/hr on Upwork marketplace bands for worldwide contracts (AI engineers, ML engineers) |
| 2 | A day | A day of workshop output | You, mostly. The meter is just larger | roughly $400 to $1,600, arithmetic from rung 1, not a surveyed rate |
| 3 | An owned document or session | The document, workspace or recording, in your accounts | The vendor. Fixed price, fixed scope | $149 AI Company Brain Webinar; $1,950 AI Company Brain Kickstart |
| 4 | A working system | One workflow in production, in your own tools | The vendor on a fixed fee; you on time and materials | Not published by Azgard, scoped on a call. Upwork comparators: $500-$1,000 chatbot, $8,000-$20,000 custom model, from about $6,000 automation |
| 5 | A month of capacity | Whatever shipped that month, plus a system kept current | Shared. You carry the standing cost, the vendor carries delivery | Not published. Structure only: Guardian 8-12 hrs/mo, Partner 15-25, Embedded 30-50, minimum three months, flexes on 30 days' notice, unused hours do not roll over |
Rung 5 goes wrong in both directions. A retainer bought before anything is built is a subscription to advice. A retainer skipped after a system ships is how a working build rots, because someone has to own it after handover and that person needs a budget line.
Rung 0 exists because most American small businesses start from zero, and at zero the honest first purchase is not a consultant. The Census Bureau's Business Trends and Outlook Survey, published May 2026, put US AI use at 17% to 20% of firms: 37% among firms with 250 or more employees, under 20% among firms with four or fewer.
Buy the lowest rung that produces something you can use without them
Buy the lowest rung of the Azgard AI Spend Ladder that produces something you can use without the consultant. Ask any AI consultant for the price of the smallest thing they will sell you, and whether you still own the output if you never call them again. A vendor whose smallest sellable unit is a five-figure discovery is asking you to fund their qualification process.
An hourly rate says who owns the vagueness. Two fears recur almost verbatim in Azgard's own harvest of buyer language: that a vague business problem just becomes an expensive project, and that the money buys a roadmap deck or an agent nobody uses after two months. Both are the same fear: paying for discovery and receiving discovery.
The fix is sequence, not negotiation. Buy the smallest rung, use what it produces, then map the work before you buy anything larger. A mapped workflow with a measured baseline turns rung 4 from an estimate into a quote, and a vendor who has seen the map prices it lower, because they are no longer pricing their own uncertainty. Sometimes hourly is still the right unit: genuinely exploratory work should not be fixed-price, and a competent contractor will refuse to fix-price a fog. The test is whether the vagueness sits in the problem or in the proposal.
What is actually included in an AI consulting fee
An AI consulting fee normally covers six things: discovery, the build, integration with the tools you already run, documentation written for your staff rather than for engineers, handover, and a warranty window for defects. Ask which of the six the quote includes. Documentation and handover get dropped most often, and they decide whether you own the result.
Four things sit outside it: model API usage and token spend on your own account; seat licenses for third-party software; your own team's hours, the largest unbudgeted line in any AI project; and data cleanup, where implementations stall on incomplete records. Those licenses are also taxed differently from consulting work in some states, so keep them on their own invoice line.
What Azgard charges: the two prices we publish and the four we scope
Azgard publishes two of its six prices. The AI Company Brain Webinar is $149: a live two-hour session, recording and starter skills pack included. The AI Company Brain Kickstart is $1,950: a done-with-you build across two sessions with 30 days of email support, and you keep the workspace. Both are rung 3, a fixed price for something you own whether or not you call again.
The other four carry no published price, and inventing one would be worse than saying so. The AI Readiness Audit, Pilot Build and Custom AI System are scoped on a call, because the price is set by how many workflows are in play. The Embed Retainer publishes structure instead of a figure: Guardian 8 to 12 hours a month, Partner 15 to 25, Embedded 30 to 50, minimum three months, then month to month.
Azgard has no US entity, no US office and no US staff. You are contracting with an Australian company in Sydney, delivering remotely and invoicing from Australia. Shared hours land in your afternoon and Sydney's next morning, as the US page sets out. One published outcome: at Apex Signage, a signage manufacturer, quoting went from 3.5 hours to just over 2 hours per quote. Azgard does not publish Apex's fee, and does not multiply hours saved by a rate to imply a return. Home Grown Electrical and Inner Game Basketball are the other two named clients.
Taxes, grants and subsidies that change what an AI implementation actually costs in the US
The invoice is not the cost. US tax treatment and state sales tax can move the real number by more than the gap between two vendors' rates, and one of those rules works against hiring an Australian consultant. Azgard does not give tax advice: everything below is dated, and classification questions belong with your CPA.
Section 174A expenses a domestic build now and amortizes an offshore build over 15 years
As of August 2026, under IRS Rev. Proc. 2025-28, domestic research or experimental expenditures are deductible in the year paid or incurred, for tax years beginning after December 31, 2024. The procedure states expressly that any amount paid or incurred in connection with the development of any software is treated as a research or experimental expenditure. Research performed outside the United States is still amortized over 15 years from the midpoint of the year incurred, and the Section 41 research credit reaches only expenditures treated as domestic under Section 174A.
That cuts against hiring Azgard. If your AI build is treated as research or experimental, a US-performed build is written off this year, while the same build performed in Sydney is spread across fifteen years and earns no research credit. It is a real cost advantage for a US supplier, and no competitor pricing page mentions it.
Whether an AI engagement is research or experimental at all is a facts question, and much small-business AI implementation - configuring tools you already license, training staff, documenting a process - is more likely an ordinary deductible business expense under Section 162, in which case neither timing nor location changes anything.
Sales tax on AI consulting depends on your state and on how the invoice is worded
Most states tax only enumerated services, and consulting is not usually among them, so across most of the country an AI consulting fee attracts no sales tax. Two verified exceptions matter. Connecticut enumerates business analysis, management and management consulting as taxable, at the 6.35% general rate, with computer and data processing services at 1%, so one engagement can attract two rates depending on how it is split. Texas puts consulting on the nontaxable list but taxes data processing at 6.25% plus local rates: entering, converting, storing and manipulating a customer's data and producing reports from it, which reads like a description of an AI implementation. Texas also applies a 5% rule. At or below 5% of the contract price and not separately identified, no tax is collected; above 5% and not separately stated, the entire charge becomes taxable.
So ask for taxable and nontaxable components as separate line items before you sign. In Texas that formatting choice alone decides whether tax lands on 5% of the invoice or 100% of it, which is worth raising early if you are hiring in Austin.
The R&D credit payroll offset is a startup test, and most established firms fail it
The federal research credit's payroll-tax offset is real and widely oversold. The Form 6765 instructions let a qualified small business apply up to $500,000 of the Section 41 credit against the employer portion of Social Security tax. Qualifying takes gross receipts under $5 million for the tax year and no gross receipts for any tax year before the five-year period ending with it. A signage manufacturer or electrical contractor with a decade of revenue does not qualify, whatever a vendor's brochure implies. The $500,000 is the headline; the disqualification is the fact you need.
What is actually open in August 2026: one free course, one closed round, one bill
Federal help for US small business AI adoption is literacy and advice, not money. The Department of Labor's Make America AI-Ready course is free, live now, a week at ten minutes a day by SMS. South Dakota ran the only program found anywhere that reimbursed a consultant's discovery fee, at 50% up to $5,000, with training at 50% up to $20,000; applications closed on August 7, 2026, so treat it as a closed round and a template for what to ask your own state for. H.R.5764, the AI for Main Street Act passed the House 395 to 14 and has sat in Senate committee since January 26, 2026 with no appropriation attached: not law, not a grant program, whatever the blog posts say. State programs open and close, so ask your state's economic development office and your local Small Business Development Center what is live this quarter.
FAQ
tags: pricing ● budgeting ● vendor-evaluation

Angus McDonald
Founder, Azgard
Builds and operates production AI systems for organisations that need results, not slide decks.

