Outsourcing After Section 174A: Eleven Months to Relearn It

Contents11 sections

For twenty years the offshore conversation was a rate-card conversation. A developer in one market costs so much, in another market less, and the argument was about whether the quality gap and the coordination overhead (the part nobody line-items) ate the difference.

In July 2025 a new variable entered, and most technology leaders I talk to have yet to fold it into the model. The US tax code now treats domestic and foreign software development completely differently, and the gap is wide enough to move decisions that used to be easy.

What changed

The Tax Cuts and Jobs Act had, from 2022, forced companies to capitalize and amortize research and experimental costs rather than deduct them: five years for domestic work, fifteen for foreign. Software development was swept in explicitly. That was brutal for software companies, especially ones with no profit to shelter, and it produced tax bills bearing no relationship to the cash a business generated.

The One Big Beautiful Bill Act, signed 4 July 2025, created Section 174A and restored immediate expensing of domestic research and experimental costs for tax years beginning after 31 December 2024. Permanently, with no sunset attached.

Foreign costs kept the old treatment. Foreign research and experimentation — software development included — is still capitalized and amortized across fifteen years.

Transition provisions covered the 2022 to 2024 capitalized balances, including a retroactive amendment window for smaller businesses that has now closed, and an election for larger businesses to pull the remaining balance into 2025 or split it across 2025 and 2026. Review those with a tax advisor rather than with an article.

What the gap is worth

Work it through. Take $1,000,000 of engineering cost at a 21 percent federal rate, setting state tax and credits aside for clarity.

On a million dollars of spend, the domestic version delivers roughly $100,000 more in present-value terms; the year-one cash difference is around $200,000. Call it a 10 percent adjustment against the spend, which is a number small enough to ignore in one comparison and large enough to reverse another, depending entirely on how wide the rate gap was before anybody opened the tax code.

Against a 60 percent offshore rate advantage, 10 percent barely registers. Against a nearshore advantage of 25 or 30 percent, it takes a real bite. And in a comparison between a US contractor and a nearshore team, it can close a third of the gap before anybody has said the word quality.

One caveat governs all of that arithmetic: this is a cash-timing and present-value effect rather than a rate change. Over fifteen years the same dollar gets deducted either way. A company carrying forward losses cares much less; a profitable and cash-constrained company cares a great deal. Which one are you? Run it on your own numbers.

A second-order effect sits alongside it. Qualified research expenses feed the Section 41 research credit, which is a dollar-for-dollar credit against tax owed rather than a deduction, and qualified small businesses can apply a portion against payroll taxes even in a loss year. How Section 174A expensing interacts with that credit, including the Section 280C election, is genuinely complicated and worth modelling properly (an approximation here is where the money quietly goes).

The rest of the model

Tax is an input now. It is still a long way from the deciding one.

The variable that dominates in practice is specification cost. Offshore delivery works where the work can be specified clearly (and specified once, which is the harder half), and the further a task sits from a clear specification, the more of the rate saving disappears into clarification cycles. With an eight-hour gap, one clarification cycle costs a day.

That is why the same vendor can deliver brilliantly on one engagement and badly on the next with the same people on it. The work was different, and only one of the two engagements could be written down.

What you cannot outsource

Five things stay inside. I have watched organizations discover each of them the hard way, and the discovery always arrives later than the decision that caused it, usually at the moment the vendor relationship ends and somebody goes looking for the person who understands the thing.

Architecture and technical strategy come first. Decisions that constrain a company for five years belong to people whose careers are attached to the consequences.

Domain knowledge comes second, and it is the one that hurts. Why does the billing system carry that exception for accounts opened before 2019? Knowledge like that accumulates through exposure, and it leaves with a vendor when the contract ends, unless somebody internal was deliberately kept alongside it.

Production ownership is third. Operations can be outsourced. Accountability to your customers for an outage cannot, and somebody internal owns the pager conversation with the board.

Security posture is fourth. It stays reviewable and auditable, and it stays yours, whoever wrote the code.

Vendor management is fifth, and it is the one that catches people out. Managing a delivery partner is a real job needing real technical judgment, so a company that outsources because it lacks engineering capacity discovers it needed engineering capacity to run the outsourcing.

A company I know outsourced an entire product line to a capable offshore partner and kept no internal engineer on it. Delivery was fine for two years. Then the relationship ended; nobody inside the company could explain how a core reconciliation process worked. Rebuilding that understanding took eleven months — eleven months of senior engineers reading code that nobody had written down — and cost more than four years of the savings. The contract was fine, the IP assignment was fine, and the knowledge was never in the contract to begin with.

Contract terms worth getting right

Exit assistance is the term people skip. Negotiating it at signature costs nothing, because nobody is thinking about leaving; negotiating it during a termination costs whatever they ask for.

How AI cuts both ways

This one is genuinely unsettled, and pretending otherwise would be dishonest.

The case that AI weakens offshore arbitrage runs like this. If AI compresses the cost of routine implementation, the work best suited to offshore delivery is exactly the work being automated, and what remains is ambiguous, architectural and context-heavy, which is the work you keep close. On that reading the offshore proposition erodes from underneath.

The case that AI strengthens it runs the other way. AI raises the floor: it narrows quality gaps and it helps with language and documentation, so a smaller distributed team covers more ground. A vendor adopting it aggressively delivers more per engineer at the same rate.

Both are happening, to different tiers of work, which is why the two cases can each be argued convincingly by people looking at the same market and describing different halves of it. The commodity tier of offshore delivery is under real pressure. That much is clear. The senior tier, meaning skilled engineers in lower-cost markets working as an extension of your team, is holding and may be strengthening. So: if your offshore relationship is priced on a volume of hands, examine it, and if it is priced on capability, it is probably fine.

Restructuring a sourcing strategy on that reasoning would be premature. Part 9's evidence says productivity claims deserve measurement before they justify a structural decision, and that applies to a vendor's claims at least as much as to your own team's.

What I actually do

1. Model after tax rather than off the rate card, with real numbers and the company's real tax position, usually with the tax advisor in the room.
2. Sort the work by specification cost first, then apply a location decision to each tier rather than to the portfolio.
3. Keep an internal owner on everything outsourced: one person who can explain how it works. That is the cheapest insurance available anywhere in this article.
4. Insist on our source control and our cloud accounts, under our identity provider, every time and in every contract.
5. Price exit assistance at signature, while it is free.
6. Prefer nearshore for product work that needs real collaboration, and reserve offshore for bounded and stable scope.
7. Fund vendor management as an actual role with an actual name on it. It stops being a real job the moment it becomes an extra duty on a busy engineering manager.
8. Re-run the model every year. The tax law changed in 2025 and it can change again, the rates move, and so does whatever AI has just made cheap.

Glossary

References

• IRC Section 174A, as created by the One Big Beautiful Bill Act, enacted 4 July 2025
• RSM US, "The OBBBA restores favorable tax treatment of domestic R&D expenses"
• Kaufman Rossin, "Navigating Section 174 research and experimental expenditures after the One Big Beautiful Bill Act"
• Anchin, "OBBBA, The Research Credit, and Section 174"
• Holthouse Carlin & Van Trigt, "OBBBA Restores Immediate Expensing for Domestic R&E Costs"
• Bennett Thrasher, guidance on the 2025 research deduction elections
• IRS Revenue Procedure 2025-28, on elections, transition rules and small business provisions
• DORA, 2025 State of AI-assisted Software Development, on AI and delivery effects

Best next read

This closes the series. A reader who came in partway through should start at Part 1, CapEx, OpEx, and EBITDA: The Same Dollar, Two Different Stories, which frames the whole argument, or at Part 2 if the vocabulary is new.

A reader about to make a decision usually gets the most from three of them: Part 3 on what becomes an asset, Part 5 on where the workload lives, and Part 7 on what the company is actually committed to.

The full series

This series is an introduction for technology leaders. It is accounting, tax and commercial context rather than accounting, tax or legal advice. Tax elections carry deadlines and contract terms are jurisdiction-specific. Model it with your CFO, your tax advisor and your counsel before you commit.