The US spent billions on border surveillance. Why can’t it catch people before they die?

An MIT Technology Review investigation found more than 1,050 migrant deaths within the range of US border surveillance towers between 2015 and early 2026, including more than 110 near Anduril autonomous AI towers since 2021. The investigation alleges persistent detection and response failures across systems operated by CBP, while finding little formal auditing of whether towers identified people in distress or why agents did not respond. Despite these findings, CBP plans to spend $1 billion on 1,497 additional towers by 2034, supported by $2.7 billion in border-technology funding approved in 2025; Anduril's tower contract is worth up to $1.1 billion and General Dynamics received an AI-tower order worth up to $115 million.

Analysis

The investable implication is procurement rotation rather than a collapse in border-tech spending. Detection hardware is becoming commoditized while the economic bottleneck is alert triage, dispatch, audit trails, and measurable mission outcomes; vendors able to bundle software, maintenance, and accountable workflow integration should gain share. GD's newer AI-tower offering has a credible opening as agencies seek a second supplier and an upgrade path from unreliable legacy equipment, while ESLT faces a longer-duration risk that installed fixed-tower systems are treated as replacement candidates rather than expandable platforms.

Near term (days to 3 months), the article is unlikely to move GD or ESLT materially because border surveillance is immaterial to consolidated earnings and appropriations remain supportive. Over 6-18 months, the more important risk is that oversight converts from a reputational issue into procurement requirements: independent accuracy testing, incident retention, response-time SLAs, and payment tied to verified outcomes would compress hardware margins and favor incumbents with service capacity. The contrarian view is that operational failures may increase—not reduce—spending, since policymakers can frame them as a staffing, command-and-control, and system-integration deficit; a broad defense-sector short is therefore poorly targeted.

PLTR is a watch item, not a direct beneficiary: the apparent gap is a decision layer that prioritizes alerts and allocates field resources, but there is no disclosed contract linkage or evidence that its platforms are being evaluated. The thesis is falsified if CBP pauses tower awards, redirects funds toward physical barriers/personnel, or adopts mandatory performance standards that materially delay deployments; conversely, a multi-vendor IDIQ emphasizing data fusion, dispatch, and auditability would validate the software-and-services angle.

Key Decisions for Investors

• Maintain a modest 6-12 month relative-value bias: long GD versus ESLT only as a small satellite pair, targeting 10-15% relative upside if GD converts its AI-tower foothold into follow-on awards; stop out if ESLT wins the next major CBP modernization tranche or GD fails to disclose meaningful border-tech backlog.
• Do not short GD on this reporting alone: border surveillance is too small versus GD's broader defense portfolio, and a compliance-driven refresh cycle could be incrementally positive for its sensors, integration, and sustainment revenues.
• Set procurement alerts for CBP tower IDIQs, task orders, and any DHS performance-standard rulemaking over the next 1-3 months. Upgrade GD only on evidence of recurring software/maintenance content or cumulative awards large enough to affect segment backlog; otherwise treat contract headlines as non-material.
• Avoid using PLTR as a sympathy long until a named DHS/CBP award, partner announcement, or disclosed pilot establishes revenue attribution. A generic AI-border narrative is insufficient given the absence of a demonstrated procurement link.