The GDPNow Q3 forecast falling from 6.2% to 4.0% has grabbed the attention, but the number that actually tells you something about the state of the economy is real final sales, now tracking at 2.3%. That is the figure worth watching, and it is down sharply from 4.2%.
Why the Headline Nowcast Misleads More Than It Informs
The consensus narrative around Atlanta Fed GDPNow tends to anchor on the baseline GDP number. Right now that number is 4.0%, still elevated, and liable to produce a round of reassurance that the US economy is running hot. The problem is that the gap between 4.0% and real final sales of 2.3% is almost entirely accounted for by Change in Private Inventories (CIPI), which contributes 1.73 percentage points to the current nowcast. Inventories net to zero over time. They are not a signal of underlying demand; they are timing. Strip them out and the picture is considerably less exciting.
Real Final Private Domestic Sales, at 2.7%, down from 5.0%, reinforces the point. The economy the GDPNow headline describes and the economy households and businesses are actually running are not the same thing. The next GDPNow update is due on 26 August, by which point more data will have been digested, but the direction of travel since the initial 6.2% print is unambiguous. A drop of 2.2 percentage points in real final sales since the opening estimate is not noise.
The AI Inventory Distortion and What the Import Data Reveals
Within the nowcast, nonresidential equipment contributes 0.40 percentage points and intellectual property 0.38 percentage points. There is, as MishTalk notes, no clean method to extract every AI-related component, but those two lines together represent a substantial portion of what is inflating the headline figure.
The AI distortion runs deeper than the nowcast alone. Yardeni QuickTakes points out that during the quarter, AI-related imports increased sharply, outpacing the large increase in US exports of crude oil and petroleum products. Imports subtract from GDP in the national accounts. So a surge in AI hardware imports depresses the headline GDP figure even as it reflects genuine domestic investment activity. The implication is that the GDP number both overstates and understates the AI effect simultaneously, depending on which line you are reading.
That said, Yardeni QuickTakes also offers a partial counterpoint to the gloomier reads. Real final sales to private domestic purchasers is tracking at 2.9%, up from 1.7% in Q1-2026, a reading that suggests the domestic demand picture, once stripped of trade-related distortions, is not collapsing. Consumer spending growth is projected at 2.0%, a rebound from 0.5% in Q1. Neither figure is spectacular, but neither supports a straightforward recession call either.
The consensus may be overweighting the headline nowcast slide as a harbinger of demand collapse when the more granular data points to something more specific: an economy warped by the scale and speed of AI-related capital flows, in both directions across the border.
The AI Contribution Numbers That Should Unsettle the Bulls
The data on AI’s share of real private fixed investment tells its own uncomfortable story. Through 2025 Q1 to 2026 Q1, AI-related spending accounted for between 121% and 127% of the change in real private fixed investment, meaning the rest of the investment basket was actually contracting, with AI spending more than covering the gap. In 2025 Q4, that figure hit 357.48%, an outlier that reflects either a genuine surge or a measurement artefact, possibly both.
Then in 2026 Q2 it dropped to 68.81%. A reading below 100% means AI investment was no longer carrying the entire load, other categories began contributing again, or AI spending decelerated, or some combination. What it does not mean is that the investment picture suddenly normalised. An economy in which a single technology theme accounts for the overwhelming majority of fixed investment growth across six consecutive quarters is, as MishTalk puts it, extremely unbalanced. The 2026 Q2 deceleration in that share is either the beginning of a healthier broadening or the start of a retrenchment. The 26 August GDPNow release will offer one more data point toward distinguishing between those two reads.
