diff --git a/holoviews/plotting/bokeh/plot.py b/holoviews/plotting/bokeh/plot.py
index c8676b32e8..b873f10085 100644
--- a/holoviews/plotting/bokeh/plot.py
+++ b/holoviews/plotting/bokeh/plot.py
@@ -15,10 +15,12 @@
                     GenericElementPlot, GenericOverlayPlot)
 from ..util import attach_streams
 from .util import (layout_padding, pad_plots, filter_toolboxes, make_axis,
-                   update_shared_sources, empty_plot, decode_bytes)
+                   update_shared_sources, empty_plot, decode_bytes,
+                   bokeh_version)
 
 from bokeh.layouts import gridplot
 from bokeh.plotting.helpers import _known_tools as known_tools
+from bokeh.util.serialization import convert_datetime_array
 
 TOOLS = {name: tool if isinstance(tool, basestring) else type(tool())
          for name, tool in known_tools.items()}
@@ -184,7 +186,32 @@ def _update_datasource(self, source, data):
             stream = self.streaming[0]
             if stream._triggering:
                 data = {k: v[-stream._chunk_length:] for k, v in data.items()}
-                source.stream(data, stream.length)
+
+                # NOTE: Workaround for bug in bokeh 0.12.14, data conversion
+                # should be removed once fixed in bokeh (https://github.com/bokeh/bokeh/issues/7587)
+                converted_data = {}
+                for k, vals in data.items():
+                    cdata = source.data[k]
+                    odata = data[k]
+                    if (bokeh_version in ['0.12.14', '0.12.15dev1'] and
+                        isinstance(cdata, np.ndarray) and cdata.dtype.kind == 'M'
+                        and isinstance(vals, np.ndarray) and vals.dtype.kind == 'M'):
+                        cdata = convert_datetime_array(cdata)
+                        odata = convert_datetime_array(odata)
+                        if len(odata):
+                            cdata = np.concatenate([odata, cdata])
+                        converted_data[k] = cdata
+                if converted_data:
+                    for k, vals in data.items():
+                        cdata = source.data[k]
+                        odata = data[k]
+                        if k not in converted_data:
+                            if len(odata):
+                                cdata = np.concatenate([odata, cdata])
+                            converted_data[k] = cdata
+                    source.data.update(converted_data)
+                else:
+                    source.stream(data, stream.length)
         else:
             source.data.update(data)