Offshore wind is a key strategic component for Europe’s energy security and for decarbonisation of energy. The EU plans to increase its capacity from 12 GW in 2020 to 60 GW by 2030, with a vision for 340 GW installed by 2050. Globally, the offshore wind capacity needs increasing from 40 GW in 2020 to 2000 GW by 2050 to meet the targets of the Paris Agreement. Currently, it takes 5 to 10 years for an offshore wind farm to become operational. Six months saving from this time in an offshore wind farm with a capacity of 1GW, by optimising the soil investigation phase, could reduce 2.2 million tonnes of CO2 equivalent in emissions by getting the GW onstream sooner, relative to the same energy generated from fossil fuels (532 tCO2e per GWh versus 6 tCO2e from offshore wind). Here we present some initial results of a physics-informed machine learning (PIML) workflow to relate cone penetrometer data (CPTU) to S-wave and P-wave velocities (Vs and Vp) at centimetric scale. The training dataset is composed of public-domain CPTU and seismic CPTU data from real investigation sites and digital analogues of Vs and Vp data of the real investigation sites obtained from analytical models of dynamic poroelasticity [1]. Tests on the PIML workflow using unseen data, show high fidelity predictions of CPTU-derived Vs and Vp. As expected, uncertainty quantification using conformance prediction shows larger confidence intervals for soils with limited trained data. Acceleration of offshore wind deployment requires optimised and more digital approaches and design strategies that can bring cost-reduction from technological upscaling and minimise environmental impact [2]. The proposed method will both inform requirements of geophysical surveys and enable early extraction of synthetic geotechnical parameters for engineering design, such as shear stiffness at small strains, and their uncertainty measure.