Nonresonant Higgs boson pair production via vector boson fusion is searched for in the final state where each Higgs boson decays to a b quark-antiquark pair, and both Higgs bosons are highly Lorentz boosted so that they can be reconstructed as two large-radius jets. The data collected by the CMS experiment in 2016-2018 from proton-proton collisions at a centre-of-mass energy of 13 TeV is used, corresponding to an integrated luminosity of 138~\text{fb}^{-1}. A novel multivariate classifier based on graph neural networks, ParticleNet, is applied to identify the jets that contain the Higgs boson decay products. No excess is observed compared to the background-only expectation, and upper limits are set on the product of the nonresonant Higgs boson pair production cross section via vector boson fusion and the branching fraction into . The strength of the coupling between a pair of Higgs bosons and a pair of vector bosons relative to the standard model coupling is constrained within the range at confidence level when all other Higgs boson couplings are assumed equal to their standard model values.