Big Tech's AI Playbook: Between Hype and Reality

After OpenAI burst onto the scene out of nowhere, plenty of commentators wondered whether the old-guard tech giants were about to be disrupted. But as I noted in my last column, the giants of the social-media era — I mean the likes of Meta and Google — haven't fallen behind much at all in the AI era. If anything, they've caught up fast and staked out every track that mattered. By now, the positioning of several of the major internet players has largely taken shape. Let's trace just how far each of them has actually gotten over the past few months.

The Chasers: Meta Lights Up the Field with Open Source, Google Stays Cool

The more interesting story isn't OpenAI, the frontrunner — let's start with the chasers, Meta and Google. Cutting to the conclusion first: Meta has gained the most, while Google is playing it safe and steady. Both companies can now credibly say they've dodged the risk of being crushed by OpenAI. At minimum, they can claim: whatever you have, I have too — and I've got some things you don't.

Start with Meta. It bet aggressively on open source and has emerged as this round's surprise winner — the network effects that open source generates shouldn't be underestimated. In a sense, Meta arrived late but got there first. Its product lineup is thinner than the competition; compared with an already-successful OpenAI or a Google sitting on a deep moat, Meta simply didn't have as strong a hand. So it went bold and open-sourced its models for everyone to use. And the move actually worked.

I'd guess PC makers are grateful to Meta too. If you're running an open-source model at home — "brewing your own elixir," as the saying goes — and you want results fast, you need to upgrade your machine. Plenty of light users who don't want to pay for cloud services try out today's small AI apps and quickly discover their graphics card just isn't cutting it, so they end up shelling out for a better GPU first. (You can just picture Nvidia's Jensen Huang grinning ear to ear after Meta open-sourced its models.)

You might dismiss this as chump change, a demand that isn't really real. But didn't the gaming-PC industry also start out as a tiny niche before becoming a major revenue source for computer makers today? DJI, the world's top drone maker, started out selling components to hobbyists too, and now consumer drones are a massive industry. So if the trend holds and more and more people get into open-source models, these AI enthusiasts could drive a whole wave of hardware upgrades. Speaking as one of those enthusiasts myself, I've sat in front of my computer waiting for results and had to stop myself, more than once, from bolting out the door to buy a brand-new machine. I feel this one deeply.

MediaTek and Qualcomm have both recently thrown their weight behind Meta's Llama 2, and will be rolling out corresponding products built around the open-source model. So this open-source trump card has proven genuinely powerful.

Google, meanwhile, is staying cool as a cucumber — after all, the Transformer architecture came out of a Google paper in the first place. Its Bard model has kept getting more polished since launch and is now built into a number of Google products; ask it to whip up a slide deck and it's done in no time. Like Microsoft, Google owns a huge stable of end-user products that everyone already uses. Case in point: no matter how much Nvidia's Jensen Huang demos onstage about what AI can do, when you actually want AI to write your report or build your slides, you're still opening up Google's online tools or Microsoft Office to do it.

So Google's AI only has to hold the last mile. It looks like they have a genuinely comprehensive integration plan in place — as I've said before, you barely need to learn anything new; the features just quietly get rolled out and handled for you, one by one. Google's products are also remarkably stable, and once you've built the habit of using them, breaking away to learn something else is more trouble than it's worth. So it looks like Google's moat is doing exactly what it's supposed to do.

The Frontrunner: OpenAI's Path Is Viable — Don't Buy the Bankruptcy Rumors

Last but not least, OpenAI. A number of Taiwanese outlets recently picked up a story claiming OpenAI would go bankrupt in 2024, and no shortage of people online debated it as if it were gospel. The story basically came from someone calculating OpenAI's server costs, multiplying by its enormous user base and running the monthly burn — and, well, the number came out huge. Oh no, they said, at this rate OpenAI will run out of money by 2024 and go under.

Let me give you the answer up front: not a chance. If OpenAI was ever at real risk of burning through its cash and folding, that window was earlier — back when multiple technical approaches were still competing for dominance. Now that everyone agrees the Transformer architecture is the winning path, and ChatGPT has north of a hundred million users, if OpenAI were somehow to go under, there'd be no shortage of people lining up to bail it out. If a company can go bankrupt after proving its technical approach works, then there's no such thing as an "AI era" to begin with. On top of that, OpenAI's chief Sam Altman is the former head of Y Combinator, the world's top startup incubator, with more successful companies to his name than you can count. If that's not about as top-tier as Silicon Valley founders get, I don't know what is.

These days, news stories often start out sounding like a comedy bit, but the more people pile on and discuss it, the more it starts to feel like real news. Still, this particular story reflects two genuine things. First, according to third-party estimates, OpenAI's user numbers have dipped somewhat — people are now pointing to a decline (July's figure showed a 12% drop). Second, it reflects real anxiety as other tech giants start catching up.

And honestly, why would you be losing sleep over whether the world's current number-one tech company is going to run out of cash and collapse? From an investing standpoint, software and internet businesses are a symbiotic indicator for phone and PC hardware across the board. Looking ahead, a few scenarios seem possible. One is that if Meta's strategy turns out to be closer to where the future is headed, demand for hardware upgrades will be that much bigger.

If the world instead moves down the Google-and-Microsoft path, things will look a lot more like today — you won't really need to train your own model. But because AI is such a major new trend, both of these logics will likely coexist side by side for a long time to come.