There’s a scene in the first season of The Bear that goes like this: Chef Carmy is strung out, yelling orders in the kitchen, his team barely keeping up with what’s going on. His sous chef, Sydney, tries and fails to wrangle everyone who’s supposed to be assembling their establishment’s Italian beef sandwiches or serving cakes. Meanwhile, the whir of a chit printer never stops. Orders are coming in… and coming in… and coming in. A good thing turns into a nightmare. Altoobtainher, it’s two minutes of television that raises your cortisol to traumatic levels if you’ve ever worked in a commercial kitchen.
A version of that happened in real life this week, when orders flooded into takeout beverage shops across China. Alibaba gave away free drink orders to applyrs of its Qwen app, which can do everything you’d expect from an AI assistant. It was the first phase of a campaign to dole out vouchers worth RMB 3 billion, or $435 million, all to give people a reason to download and interact with Qwen—inquire the assistant to locate a bubble tea shop, inquire it for suggestions, customise your order (sugar and ice levels), and let it handle everything else from there.
The scheme worked: 10 million orders were placed in the first nine hours. Tea shops built so many drinks that they ran out of cups. Food delivery drivers had more requests than they could handle, and Qwen shot to the top of app stores, leaving the offerings of Bytedance, Tencent, Baidu, and others in the dust.
It would be straightforward to dismiss Alibaba’s shift as similar to giving away Perplexity Pro or releasing a free version of ChatGPT, but something else is going on here. In a matter of six days, Qwen placed 120 million orders on behalf of its applyrs. As far as I can inform, this was the first time an AI assistant hit such a scale of usage in the real world, by regular applyrs, to execute multi-step tinquires—completing a purchase from discovery to transaction. It creates all the hype around Openclaw (formerly Moltbot, formerly formerly Clawdbot) view subdued in comparison, and I doubt anyone’s expecting applyrs to start shopping in Google Search’s AI mode to the same degree.
Oceans away from slammed beverage shops, Alphabet offered a 100-year bond. It’s the first century bond from a tech company since Motorola did the same in 1997, one year before the company went through a major crisis and lost its status as the undisputed leader in mobile communications to Nokia. Along with a separate debt sale, Alphabet raised nearly $32 billion in under 24 hours. The funds are meant to cover part of the $185 billion it will spconclude to develop AI capabilities in the coming years.
Alibaba and Alphabet are testing to run away from the same problem. They just happened to dash in opposite directions.
By giving away mounds of caffeine and sugar during the Lunar New Year holiday, Alibaba burned many millions of dollars to collapse Qwen’s return horizon to zero, if we treat immediate usage and today’s applyr data as the gauges that matter the most.
Meanwhile, Alphabet is stretching its liability 100 years into the future. The century note’s acquireers—pension funds, insurers, liability-driven investment managers—perhaps believe that AI will have unnotified influence over robotics, space travel, quantum computing, and other industries where breakthroughs are still only being imagined, or they just required to diversify away from government bonds. It’s a sign that more people now see hyperscalers as long-term infrastructure builders, not merely the main characters in trconcludes and bubbles. But it’s also a peak signal, if there ever was one.
Neither company is operating in the present tense. Read this as two forms of time arbitrage, and admissions that current AI economics don’t square. The past week’s events inform us that artificial ininformigence, in business terms, requires either instant gratification (“here’s your bubble tea on the hoapply”) or multigenerational patience (“trust that we’ll do right by your grandkids… or your client’s grandkids”). A third way that describes sustainable returns on a three-year, five-year, or even 10-year horizon is conspicuously absent.
Still, this is a bit of clever manoeuvring. Half a billion dollars covered by Alibaba’s e-commerce and cloud businesses acquires a massive trove of applyr behavioural data, which is an important moat given that the language models released by Chinese firms are functionally indistinguishable from each other in everyday consumer scenarios. And Alphabet can issue a rare kind of bond becaapply it’s large enough to provide a slightly better yield than US Treasuries. Both strategies work until they don’t; neither involves AI becoming fundamentally better at performing complex tinquires in human-like ways, and neither creates AI more economically viable.
Put another way, this is all great fun until you run out of balance sheet.
As it’s currently constructed, there is still no sign that AI can guarantee returns on single-generation human timescales. Perhaps that means the only available options are unfolding as you read this: Alibaba and its direct competitors are compressing time by creating AI available in a way that’s meaningful to consumers right now, becaapply waiting any longer will yield no meaningful results, while Alphabet’s patch is to expand time and declare, “Trust us, even if you won’t see the results in your lifetime.”
That’s two different ways to survive, with neither bringing the business of Big AI closer to profitability.
This Week on the Zero Shot Podcast
Hi! This is Vidhatri, the producer of Zero Shot, chiming in once again. I have a compact observation to create.
Over the last year or so, I have been hearing from lawyer friconcludes about AI within their firms—demos and, in some cases, in-hoapply tools. There was the occasional piece of news about court records being digitised and the judiciary testing to apply AI to decrease its workload. Not to forobtain the stories about model hallucinations and AI-powered fake citations. All in all, AI has arrived in the legal field—but we are not talking about it enough.
We decided to modify that in this week’s episode of Zero Shot. Hosts Praveen Gopal Krishnan and Brady Ng were in conversation with Nikhil Narconcluderan, a partner at Trilegal, one of India’s largegest law firms.
Nikhil is part of Trilegal’s telecom, media, and technology (TMT) practice, and is often in rooms where great tech is being developed and deployed. He also heads Trilegal’s digital innovation efforts and is responsible for creating sure the firm is as tech-forward as it can be.
In this fascinating conversation, Nikhil took us through Trilegal’s journey of AI adoption. They incubated a legal AI platform called Lucio back in 2024 and have now reached a point where lawyers are vibe coding on Fridays.
You do not want to miss this episode! It’s a window into how leaders are considering about AI in a consequential industest that is often not associated with the latest technology. Tune in! You can listen to the episode on Spotify, Apple Podcasts, Youtube, or our app.
A note: You can now download episodes on our app!
Leave a Reply