Google Gravity Balloon -

Google Gravity Balloon -

[ V = \frac{m_{air}}{\rho_{strat}} \approx \frac{30 \text{ kg}}{0.088 \text{ kg/m}^3} \approx 340 \text{ m}^3 ]

1. Introduction: The 95% Problem In 2011, Google X (now X Development) proposed a radical solution to a persistent economic reality: while satellites offered global coverage but were expensive and high-latency, and cell towers offered high bandwidth but were geographically limited, nearly 95% of the world’s population lived within range of a cellular signal—yet only half were connected. The problem wasn't coverage; it was economic viability in rural and remote regions. google gravity balloon

The "Gravity Balloon" (a nickname derived from its buoyancy-based altitude control) was not a balloon in the party sense, but a operating in the stratosphere—a realm colder, drier, and more violent than most aircraft ever encounter. 2. The Physics of Floating Against Gravity To understand Loon, one must first understand the stratosphere (10 km to 50 km altitude). Below 10 km, weather dominates: wind shear, turbulence, precipitation. Above 20 km, the atmosphere is stable, with predictable zonal (east-west) wind bands. However, at 20 km, air density is only 7% of sea level. The "Gravity Balloon" (a nickname derived from its

Loon’s envelope used helium. To lift a 15 kg payload (electronics + batteries) plus a 15 kg envelope, the balloon required displacing ~30 kg of air. At 20 km altitude (pressure ≈ 50 hPa), the volume needed is: Below 10 km, weather dominates: wind shear, turbulence,

Rather than a sphere, Loon used a lobed structure (like a pumpkin) with a tendon network. This shape allows pressure-induced stress to distribute along the seams, not the film. The film itself was a 0.076 mm thick co-extruded polyethylene with a specific UV-resistant additive. The seams were reinforced with load tapes.

The optimization problem: maximize the number of user-hours connected given constraints on battery (solar recharge rate), wind prediction error, and balloon longevity. This became a partially observable Markov decision process (POMDP) with >10^6 state variables.