Left: a vibrating bond can never have zero energy, its lowest level is E₀ = ½hν and its position is smeared out. Right: close two mirrors and the vacuum between them pushes less than the vacuum outside, so they are pressed together.
The science
Oscillator. A mass μ on a spring obeys En = (n + ½)·ħω, ω = 2πν. The ground state is not at E = 0: the uncertainty principle forbids a particle to have both exactly zero position and momentum, so E₀ = ½ħω remains. Wavefunctions are ψn ∝ Hn(x/x₀)·e−x²/2x₀² with length scale x₀ = √(ħ/μω), and ⟨x²⟩ = (n + ½)·x₀². The dots are random draws from |ψ|² (the actual, only meaningful probabilistic picture).
Casimir force. Summing the ½ħω of every electromagnetic mode between two ideal, parallel, perfectly conducting plates and subtracting the same sum for unbounded space leaves a finite energy per area E/A = −π²ħc / (720·d³). Its derivative is an attractive pressure P = π²ħc / (240·d⁴): about 1.3 mPa at 1 µm and about one atmosphere near 10 nm. It was measured by Lamoreux (1997) and later experiments to a few per cent.
Honest limits. Zero-point energy is the lowest energy a system can have, so it cannot be extracted to power anything; the Casimir effect is a difference between configurations, not free energy. Real metal plates are not perfect (the force is smaller at gaps above ~100 nm and depends on temperature and roughness; Lifshitz theory is needed). Large "vacuum energy density" claims in popular articles come from adding up modes to an arbitrary cut-off and are not predictions of this model. The picture of mode waves is schematic and not to scale. The CO-like preset uses a typical stretching frequency of about 64 THz and reduced mass 6.86 u as an illustrative value.