Jason was reacting to the FCC’s decision in December 2023 to deny Starlink roughly $885–886M in Rural Digital Opportunity Fund (RDOF) subsidies, which were intended to finance future rural build‑outs, not to pay for existing end‑user subscriptions. That ruling reaffirmed an earlier 2022 revocation of the provisional award. It did not cut off service to a cohort of current Starlink customers; it simply meant Starlink would not receive that public money to expand to 642,000 planned rural locations.(irepod.com)

Separately, the main U.S. household internet subsidy—the Affordable Connectivity Program (ACP)—ended on June 1, 2024 after connecting over 23 million low‑income households.(congress.gov) However, multiple independent consumer and Starlink‑focused sources agree that Starlink either never meaningfully participated in ACP or was at most briefly/nominally listed and then removed; in practice, Starlink customers did not receive ACP discounts at scale.(highspeedinternet.com) That means there is no well‑documented, sizeable group of “rural customers whose government‑subsidized Starlink service was canceled” to study.

Where we do have data is for non‑Starlink ISPs: FCC surveys and independent analyses found that when ACP funding lapsed, roughly three‑quarters of participating low‑income households expected to change or drop their internet service rather than keep paying full price, and providers like Comcast in fact lost tens of thousands of ACP‑linked broadband customers when the subsidy ended.(route-fifty.com) Meanwhile, Starlink’s global subscriber count rose sharply from about 2.2 million in December 2023 to around 8 million by late 2025, with a large share of its base in rural areas, but those aggregate figures don’t tell us who subscribed or whether they were formerly government‑funded users.(en.wikipedia.org)

Because (1) the specific population Jason describes (rural users whose Starlink was canceled due to a U.S. subsidy program) is not clearly identifiable or sizable in the real data, and (2) we lack provider‑level evidence on what such users actually did afterward, his prediction about that group’s behavior cannot be directly evaluated from available information. Hence the outcome is ambiguous rather than clearly right or wrong.