Such graphs are studied a lot in coding theory and in the theory of association schemes. In paricular the eigenvalues can be explicitly written down  in terms of Krawchuk polynomials $K_k(u)$. 

You can find this e.g. in Bannai  & Ito "Association Schemes I": https://www.amazon.com/Algebraic-Combinatorics-Association-Schemes-Mathematics/dp/0805304908

I never saw inertia of these graphs mentioned anywhere, but it looks easy to figure out.

To give you an idea, the eigenvalues of the relations of the Hamming scheme $H(n,2)$ (that is, the $n+1$ adjacency matrices $A_j$ of graphs $\Gamma_j$ ($0\leq j\leq n$) partitioning $K_{2^n}$, so that two vertices of $\Gamma_j$ are adjacent if the Hamming distance between them is $j$, may be compactly presented (discounting multiplicities) as an $(n+1)\times (n+1)$ matrix $P$,
with $j$-column containing the eigenvalues of $A_j$, and $j$-th row containing the eigenvalues corresponding to the $j$-th common eigenspace of $A_j$'s (the latter works as $A_j$'s pairwise commute).

In general, the entries of $P$ are given by
$$
\newcommand{\kk}{k}
\newcommand{\uu}{u}
P_{ku}=K_{\kk}(\uu)=\binom{n}{\kk}\sum_{j=0}^k
\frac{(-\kk)_j(-\uu)_j 2^j}{(-n)_j j!}=\binom{n}{\kk}\sum_{j=0}^k (-1)^j 2^j
\frac{\binom{u}{j}\binom{k}{j}}{\binom{n}{j}}
$$

To get the eigenvalues of $H_2^n(d)$ you just need to sum up the last $n-d+1$ columns of $P$. (The multiplicities are given by the binomial coefficients, also this is what you get in the 1st row of $P$).

-----------------------

Here are the experimental results; I [coded up this][1] in [Sagemath][2] and computed the number $p(d)$ of positive eigenvalues of $H_2^{2d}(d)$ for 
$1\leq d\leq 18$. Here is what I got

```[1,6,36,136,496,2016,8256,32896,130816,523776,2098176,
 8390656,33550336,134209536,536887296,2147516416,8589869056,34359607296]
```

While the usual interface does not produce a match, [Superseeker][3] does tell me a lot, e.g. it says that 


[Guesss][4] (a program by Harm Derksen) suggests that the generating function  $F(x)$
may satisfy the following algebraic or differential equation:

$$x^2+1/8 x+1/16+(x^3-1/4 x^2+1/4 x-1/16)F(x) = 0$$

as well as that the exponential generating function for this sequence is
$$2 \exp(4 x) - \sin(2 x) - \cos(2 x).$$

Thus, conjecturally, this is the answer for the number $p(d)$ of positive eigenvalues; there are no 0 eigenvalues, and the number of the negative ones is just $2^{2d}-p(d).$

---------------------

But in fact it's much easier to state, namely, conjecturally
$$
\frac{2^{2d}-2p(d)}{2^d}=
\begin{cases} 
1 &\text{if $d\mod 4\in\{1,2\}$ and}\\
-1 &\text{otherwise.}
\end{cases}
$$
  [1]: https://gist.github.com/dimpase/9f1c68cbc1d64d447f62efdca9b3ce7b
  [2]: http://sagemath.org
  [3]: https://oeis.org/ol.html
  [4]: https://oeis.org/superhelp.txt