Having a separate research division being an anti-pattern is an interesting topic! I remember getting into a related discussion almost a decade ago with a professor who left academia to join Google, his point being product-driven R&D was strictly "better" than "bluesky" R&D because (IIRC) the work is more directly related to market needs.
My contention was that this ignores the transformative potential of long-range theoretical research. For instance, somehow very few consider Xerox PARC to be an anti-pattern.
From what I hear in the last few years even MSR has changed its ways to steer its research more in line with needs of product divisions, and I actually consider that a loss. Who knows what paradigm-shifting inventions like GenAI are being steered away from?
> Look at how much Microsoft spends on R&D for the last 25 years, compared to the amount Google spends, in absolute terms and as a fraction of revenues.
Hmm, at the risk of relying on sycophantic bots, AI overviews suggest most recently Microsoft spent 13.2% of revenues vs 14.8% for Google (and 30% for Meta!) Of course even a single % point is in the millions at their scale, but there are a ton of confounding factors including differing product margins and payscales (and CEO obsessions like Metaverse!) At least at a quick glance MSFT and GOOG seem comparable.
The problem here is how "R&D" is defined. Unfortunately, even day-to-day product development is lumped in with R&D. I've done "R&D" in academia, private research firms, and big tech, and they are all poles apart. "Actual" R&D is very researchy, often based in discovering new aspects of reality, whereas "product" R&D is just regular product development. Which could be considered discovering new aspects of the market I suppose. They are both valuable but on very differnt timelines.