According to Techmeme, SoftBank has completed a $22.5 billion investment in OpenAI, finalizing the up to $40 billion commitment it first announced back in March. This latest cash infusion gives the Japanese investment giant an approximate 11% stake in the leading AI company. The news comes alongside commentary from tech observers like Gergely Orosz, who points out that Meta, despite being incredibly active in AI research this year, still lacks a true market-leading AI product. Meta’s recent acquisition of Manus AI has sparked debate, with Orosz and others recalling Meta’s (then Facebook’s) 2013 purchase of the hot developer tool Parse, only to shut it down four years later. The immediate outcome is a clearer picture of SoftBank’s huge bet, but a cloudier one for Meta’s AI ambitions heading into 2026.
Meta’s Checkered Acquisition History
Here’s the thing: when Gergely Orosz brings up Parse, it’s a legit warning shot. Everyone gets excited about a big tech giant snapping up a promising startup. But how often does that actually result in a world-changing product from the acquirer? Parse was a mobile backend darling, and Facebook bought it… and then did basically nothing before killing it. It’s a pattern. Look at Oculus—sure, it’s still around, but has it truly revolutionized Meta’s core social fabric? Or did it just become a costly vehicle for a metaverse vision that hasn’t landed? So when Meta buys Manus AI, a smart observer’s first reaction shouldn’t be “Wow, game on, ChatGPT!” It should be, “Okay, but what’s the actual plan here?” Will this be integrated, or will it languish and die on the vine in a few years? The historical data isn’t encouraging.
The Real AI Product Gap
And that leads to the core issue. As noted, Meta is shockingly one of the few giants without a true, standalone AI product in the market. Google has Gemini (for better or worse). Microsoft has Copilot woven into everything. Apple is pushing Apple Intelligence. Even startups are launching new models weekly. Meta has Meta AI, but let’s be honest—it’s an assistant buried in their apps, not a destination. They’ve released great open models like Llama, which is fantastic for the ecosystem and researchers. But that’s not a consumer product. It’s like they’re building the best engines but haven’t bothered to build a car anyone can drive. The Manus acquisition suggests they might be realizing this gap and trying to move beyond social media features. But is buying a startup the answer, or is it a sign they can’t build it themselves?
SoftBank’s All-In Moment
While Meta’s strategy seems scattershot, SoftBank’s is brutally simple: go huge on the perceived leader. Dropping $22.5 billion at once isn’t a dabble; it’s a conviction that OpenAI will be the dominant force. This isn’t a portfolio diversification play. It’s a massive, concentrated bet that the AI platform of the future is being built at OpenAI right now. It also raises the stakes for everyone else. If you’re a competitor, you’re not just fighting OpenAI’s tech and talent anymore. You’re fighting against that war chest and the patience it buys. For SoftBank, after some spectacular missteps in recent years (WeWork, anyone?), this feels like a desperate need to anchor its legacy to the one trend that everyone agrees is real. The question is, what happens if OpenAI stumbles? There’s no Plan B here.
What 2026 Could Look Like
So what does all this mean for the near future? Orosz speculates we might see a complete relaunch of Meta AI in 2026, a real attempt to challenge ChatGPT head-on. That’s one path. The other, darker path is that Manus and other assets just become bargaining chips or talent acquisitions, never fully integrated. Meta might just be amassing AI “options,” hoping the rising tide of open-source models they help create lifts their own social media boats. But that seems like a timid strategy for a company that needs a win. Meanwhile, SoftBank’s capital means OpenAI can ignore the noise and the costs for a long, long time. The pressure isn’t on them to monetize tomorrow. The pressure is now squarely on everyone else—especially Meta—to prove they can build something people actually choose to use, not just something that comes pre-installed in an app. Can they?
