* You're modifying GPL'd software, which means you're required to release the source code including your modifications.
* Pushing the patches upstream reduces the costs to your business in the long-term, because it avoids the need to maintain separate branches for your own changes.
IMO, the second reason is a much bigger win for businesses, and exactly the reason permissive licenses (BSD, Apache 2.0) are getting more and more popular. Permissive licenses allow businesses to collaborate on code which is going to be common within and across industries (reducing development costs and time-to-market) while allowing them to keep some other parts of their code private.
The first is an unquestionable reason; it's a lot easier to get him to agree if it's a legal requirement. So even though the real reason why it benefits the company is the second, the first provides the leverage to tip the scales in favor of releasing the code. If the code were BSD licensed, we would likely never ship it upstream, because of his fear that our competitors will take advantage of it.
And there's an additional, third factor in play: if you release it under the GPL, you don't have to worry about your competitors making a proprietary fork of it, thus taking advantage of your work while keeping their own secret. So that further tips the scales. Sure, with GPL you may be required to release your modifications, but at least you know that by doing so, you won't be giving your competitors a major advantage if they then build something proprietary on it.
And finally, in many cases with GPLed licenses, you can keep your own code proprietary. Just write a separate application, rather than linking it with the GPLed components. That's how most of what my company does works; we sell a system based on Linux and many other GPLed components, but we tune it to the hardware we sell, and have our own proprietary management tools on top of it that make it integrate well into our market. This is why the LGPL or more permissive licenses are generally preferred for libraries (with very few exceptions, such as readline), which are designed to be reused regardless of the license of the application, while the GPL works so well for components like the kernel, system services, and so on.
As a developer who works with a boss who has a somewhat equivocal position on open source software, I'm glad for the GPL requirements on the software that we modify and ship. He likes to use open source software, and in theory likes to pay lip-service to contributing change upstream, but when push comes to shove he's always worried that we'll let competitors get a jump on us. With GPL licensed software, I can always make the argument that we have to release the source code anyhow, so we might as well push the changes upstream and avoid the pain of maintaining a fork. With other software, well, it's always somewhere far down on the priority stack to actually release the code, and so we wind up maintaining patch stacks for years and shipping with major security vulnerabilities because we're afraid to update as it will take a lot of work to port our changes forward.