You're making up numbers. We've had billions of transistors on chips for some time now. The XBox One's main chip has five billion transistors. And that's just one chip. The Titan supercomputer has nearly 200 trillion transistors.
If the transistor doubling time remains about the same, you can equate any number of transistors you like to a neuron and Kurzweil's prediction still won't be off by much. Such is the nature of exponential curves. Sophisticated objections to his predictions don't involve transistor counts.
Nobody knows how much of a neuron you need to build a brain. If you actually have to simulate it, possibly at the quantum level, then no number of transistors may be sufficient. You can probably get around that problem by not using regular transistors though. Sufficient artificial neurons might actually be easier to build - noise and interference are probably not as harmful as they are in regular computing, and may actually be beneficial.