The Empire Strikes Back — Google's Turn

A major development just happened — Google held its I/O conference and unveiled a huge wave of progress on Gemini. The long-awaited Google "Empire Strikes Back" moment has finally arrived.

I've pointed out in this column before that Google, while slower to move, has more synergistic resources at its disposal than anyone else. So it's always in the game — even if it can't be first, it can be second. But it held that second, or even third, position for a remarkably long stretch — until right around this conference, when it finally looked like the company had committed, fully, to going all in on AI. And in doing so, it seems to have forced OpenAI's hand on a string of ChatGPT updates, and forced Anthropic — the independent number one — to ship Claude 4.0.

Google Shows Its All-In Resolve, Forcing ChatGPT and Claude to Respond

One tangible metric worth watching: Gemini's coding ability has improved dramatically, and that's put real pressure on OpenAI.

Another shift: Google's video model Veo 3 is remarkably good, and it isn't expensive either. On Twitter you can already see users producing what look like genuine film clips with Veo 3 — the main limitation is that clip length is still capped. I tried making a few cute-animal videos, and everyone I sent them to was stunned; the output feels indistinguishable from professional production.

Worth noting: one of the biggest changes in Google's AI integration is that it has now merged search with AI summaries into a single interface. Granted, the quality of AI search results isn't great yet — it has a bit of a "ship it now, polish later" feel. But it's a clear signal of how Google's thinking has shifted.

Why does this matter? Because Google has always had one core problem with AI: search is just too profitable. So in the past, Google couldn't afford to get ahead of OpenAI in the AI race — what if it cannibalized its own search advertising? But now Google seems to have concluded this shift is unstoppable, and it apparently would rather disrupt its own market itself — and figure out how to monetize the disruption from the inside — than let someone else do it to them. That's a striking level of resolve.

Meanwhile, Google's core business keeps getting stronger, because it has another growth engine: YouTube. So my read is that Google's growth trajectory isn't slowing down at all — and it has an enormous additional runway in YouTube. Video data volume and video data usage are, by far, the largest datasets ordinary users generate today, and having that kind of data source in-house is a genuinely enviable position.

A quick aside here: back at the start of the year, everyone was asking whether DeepSeek would catch up. As it stands now, its usage share has fallen from 6% at the beginning of the year to 2%, or even less. Of course, DeepSeek's core advantages haven't really changed — Chinese domestic firms, with their huge manufacturing base, still lean on DeepSeek for hardware integration. And with China's massive user base — plus the fact that Meta's Llama 4 doesn't look as successful as DeepSeek — DeepSeek still holds an edge in the open-source category. So broadly speaking, it's the Chinese market and the open-source crowd still using DeepSeek, while global mainstream users remain with the original big three.

Another open question is OpenAI's Stargate project, reportedly a $500 billion undertaking. Oracle, one of the big application vendors in this space, has now fully thrown in with the OpenAI camp — they're buying $40 billion worth of Nvidia chip capacity for OpenAI's alliance to use. That solves OpenAI's compute supply problem for years to come.

It looks like the coming years will settle into something like a "Google camp vs. non-Google camp" structure. It's quite possible the non-Google camp eventually consolidates entirely around OpenAI to compete with Google. The bigger question is whether Google has a shot at overtaking OpenAI this year. The current competitive dynamic is really Google and OpenAI fighting over who gets to be the biggest AI company.

What happens to Anthropic — the smallest of the three giants — is still unclear, because it comes down to whether the market has room to support an independent number three. Right now it looks like there's enough room, but three years from now is anyone's guess. As for Meta, it finds itself in a rather odd spot: Llama 4 hasn't been a huge success, and DeepSeek is out there acting as a hedge against it in the market.

Google Takes Aim at OpenAI — This Year Should Show Who's Winning

The current competitive landscape among the big AI labs is tilting back toward closed models — the closed-model players have basically played all their cards now, nothing left in the hand. This year should be the one where the outcome starts to become clear.

But has anyone noticed something interesting? As competition intensifies, vendors that were already required to buy Nvidia GPUs are being forced to buy even more. So despite the geopolitical risk, Nvidia's performance is still holding up remarkably well. In the foreseeable near term, none of the current alternatives to Nvidia GPUs actually deliver comparable value.

Given the competitive dynamics on the software side, companies like Oracle need to start planning these procurement decisions now — you can't exactly go pick up GPUs off the shelf when you need them. In the AI space right now, the application layer is still relatively early, and shovel sellers are still doing far better than the gold miners. If you're positioning an investment portfolio, that's worth watching.