$\eta$ Carinae is composed of two massive stars orbiting each other in 5.5 years. The primary star features the densest known stellar wind, colliding with that expelled by its companion. The wind collision region accelerate particles up to relativistic energies, producing non thermal X- and $\gamma$-ray emission detected by high energy space missions and H.E.S.S.. The orbital variability of the system provides key diagnostic on the physics involved. The low-energy component, which cuts off below 10 GeV, is likely of inverse Compton origin. The high energy component varied by larger factors and differently during the two periastrons observed by Fermi. These variations match the predictions of simulations assuming a magnetic field in the range 0.4-1 kG at the surface of the primary star. The high-energy component and the thermal X-ray emission were weaker than expected around the 2014 periastron suggesting a modification of the inner wind density. Diffuse shock acceleration in the complex geometry of the wind collision zone provides a convincing match to the observations and new diagnostic tools to probe the geometry and energetics of the system. A future instrument sensitive in the MeV energy range could discriminate between lepto-hadronic and hadronic models for the gamma-ray emission. At higher energies, the Cherenkov Telescope Array will distinguish orbital modulations of the high-energy component from these of ultraviolet-TeV photo absorption providing a wealth of information constraining acceleration physics in more extreme conditions than found in SNR. Hydrodynamical shocks in the vicinity of massive stars can accelerate particles to very high energies and generate positrons.