The IRA, the CHIPS Act, and the Section 174 Surprise

Three federal changes shape the current build environment, and each works differently. Read them as separate mechanisms rather than a single bucket of 'incentives.'

The first is the set of clean-electricity credits enacted in the 2022 Inflation Reduction Act. These reward investment in, and generation from, low-carbon power. They do not pay for a data center directly. What they do is improve the economics of the generation a data center increasingly wants attached to it — the solar, wind, storage, and other projects that operators contract for to secure power and meet clean-energy commitments. When you read about a campus paired with a large power project, the federal credit is often part of why that project financed. The credits were also designed to be transferable, which widened the pool of parties who could benefit from them — a structural detail worth knowing.

The second is the semiconductor manufacturing credit created by the CHIPS and Science Act, a substantial investment credit aimed at facilities that fabricate chips. This one sits upstream of the data center rather than inside it: it subsidizes the fabs that make the accelerators, not the halls that run them. Faraday's read: its relevance to infrastructure investors is as a supply-side signal — where fab capacity is being incentivized shapes the medium-term availability and geography of the silicon everything else depends on.

The third is the quietest and, for a few years, the most painful: a change to how research and development costs are treated. Beginning with tax years after 2021, a provision commonly referenced by its section number required companies to capitalize and amortize research expenditures over several years rather than deduct them immediately. For software-and-engineering-heavy operations, that change raised near-term taxable income and pinched cash flow even though nothing about the underlying business had changed. It is the cleanest example in this course of a stroke-of-the-pen tax change with no siting consequence and a real balance-sheet consequence.

The empiricist's lesson from all three: federal policy is fragile in a specific way. Statutes get amended, credits get extended or curtailed, and transferability rules get rewritten. A model built on a federal credit is exposed to a legislature, not a county commission. Track the statute, not the summary.

Hold onto the distinction between the three mechanisms, because they fail in different ways and on different clocks. The generation credits are exposed to the durability of the enacting statute and to the rules governing how the credits can be transferred or monetized; a future Congress can narrow either. The fabrication credit is a multi-year industrial-policy commitment whose relevance to infrastructure is indirect and slow-moving. The research-cost change is the cleanest cautionary tale of all: a provision that touched no incentive program, moved no campus, and yet measurably worsened the cash flow of exactly the kinds of engineering-intensive firms that build and operate this infrastructure. Faraday's read: the lesson for an analyst is to treat federal tax exposure as a distinct risk category with its own volatility, separate from the negotiated-but-durable character of local deals. Model the local abatement as a contract; model the federal credit as a policy position that can change with an election. Confusing the two — treating a statutory credit as if it were as reliable as a signed county agreement — is one of the most common and most expensive errors in the whole field. Put plainly: model the county agreement as a contract and the federal credit as a policy position, price each for the risk it actually carries, and never let the durability of the former lull you into trusting the fragility of the latter.

The applied discipline is to tag every federal exposure in a model with its own volatility note: this credit depends on a statute that could be narrowed; this depreciation rule already changed once and could change again. Treating the federal layer as a set of policy positions rather than fixed parameters keeps a model honest about the risk it is actually carrying. It also sharpens forecasting, because shifts in federal tax posture are frequently telegraphed in legislative text and committee activity well before they take effect, giving the attentive analyst a genuine lead on how the market's build-or-wait calculus is about to move.