ByteDance, OpenAI, and Meta are all betting on one thing.

Have you noticed that large companies are all deploying their own AI hardware products?

At the Davos Forum, Chris Lane, the global affairs officer of OpenAI, revealed the latest news that OpenAI is progressing as planned and preparing to launch its first AI hardware device in the second half of 2026.

This news has completely stirred up the already anxious tech circle.

This means that the world's top algorithm engines are entering the hardware race. In China, recently, companies like Huawei, ByteDance, and Alibaba are all flocking to the AI hardware market.

You know, in the past decade, Internet companies have been trying to get rid of bulky hardware and pursue a light - asset software model. Now, they are collectively "reverting to their ancestors" at all costs and rushing back to the tough hardware market that once gave them a hard time.

Take ByteDance as a typical example. As a software company, its Feishu recently collaborated with Anker Innovations to launch an AI office hardware called "One Bean". Before that, people only knew Feishu for its documents, spreadsheets, and video conferencing.

But if you broaden your perspective, you'll find that this is not just Feishu's move. ByteDance, Alibaba, Xiaomi, Baidu, and even Meta and OpenAI across the ocean are all scrambling into the hardware market.

Meta is the most aggressive. According to media reports, Zuckerberg has asked manufacturing partners to double the annual production capacity of AI smart glasses to 20 million pairs by the end of 2026.

Is this another technological upsurge? However, if you look back at the pitfalls these large companies have encountered on the hardware path, you'll find how abnormal this collective influx is.

In the past decade, large companies' attempts at hardware almost always ended in failure, especially in the Chinese market. Baidu made mobile phones, ByteDance acquired Smartisan, Alibaba launched Tmall Genie, and Tencent also tried various social hardware.

What was the result? Either they left with losses or disbanded the departments. It seems that software companies' DNA is not suited to handle hardware.

For large companies to venture into hardware is like jumping into a fire pit.

But in 2026, the hardware trend has not only returned but is even more intense than before.

What exactly is going on behind this? Why are large companies willing to jump into this arduous hardware business again?

1. Why can't software companies do well in hardware?

Before discussing why everyone wants to do hardware now, we need to see why they failed in the past.

You may have forgotten that Tencent once wanted to make mobile phones. Around 2015, Tencent launched the TOS (Tencent OS) system and even collaborated with manufacturers to produce mobile phones. What was the result? There was no splash in the market, and the project was soon halted. After ByteDance acquired Smartisan, it also couldn't escape the fate of failure.

The reason why it's difficult for large companies to do hardware is that they carry the arrogance of Internet companies in their genes.

The logic of Internet companies is to quickly test and iterate. If there is a mistake in an app, programmers can stay up all night to modify the code, and users can resume normal use after an update the next day. In this world, the marginal cost is almost zero. One more user just means a little more server space.

But hardware is a completely different story.

The logic of hardware is a one - shot deal. If there is a one - millimeter error when making a mold, hundreds of thousands of shells will become waste. If you order one less sensor in the supply chain, your entire production line will have to stop. Hardware companies pursue cost control of each part and inventory turnover.

Take the smart speaker war back then. To grab users, large companies sold speakers that cost 200 yuan for 69 yuan or even less. They thought that as long as you bought their speakers, you would use their music services and buy their products.

Users did buy, but they just used the speakers as alarm clocks or radios. The hardware itself didn't make money, and the software services couldn't drive sales either. In the end, these hardware departments became a heavy burden on the company's books.

In short, software companies didn't understand the respect for hardware before. They approached hardware with the same mindset as making apps and were finally taught a lesson by the supply chain and inventory.

2. AI has changed the rules of the game

Since it was so difficult in the past, why do people think it's okay now?

The answer lies in the large - scale AI models. The most important reason is actually just two words: interaction.

In the past, it was difficult to make hardware. You had to design screens, complex menus, and figure out how users would click and press. This required strong industrial design and interaction logic. Software companies' failure in hardware was largely because they designed the hardware too complexly and couldn't handle these physical - level complex interactions.

But after the emergence of large - scale AI models, the situation has changed.

Nowadays, AI hardware no longer needs a screen or even redundant buttons. Look at Feishu's "One Bean". Its core interaction is simply listening and speaking.

Take AI glasses as another example. You say a word to the glasses, and they can understand your meaning and execute directly. This interaction method has lowered the threshold of hardware.

Software companies have found that they don't need to study how to make a perfect touch screen. They only need to do a good job on the "brain" (AI model), and the hardware only needs to provide a microphone, camera, and chip.

Nowadays, hardware has essentially become a sensor for AI.

Take the Plaud Note product as an example. It sells well because it addresses the pain point that the iPhone can't record calls. But do users really buy it just for recording? No. Users buy it for the summarization, transcription, and translation after recording.

Without a model like ChatGPT behind it, this recording pen would just be ordinary electronic waste. It's AI that gives these simply - structured hardware devices the ability to think and summarize.

Software companies suddenly realized that they don't need to compete with Apple and Huawei in terms of precise industrial design. They only need to make a simple carrier and install their AI models into it.

For Feishu, the original intention of making "One Bean" is definitely not to make money from selling hardware, but to let Feishu's AI go beyond computers and mobile phones. If you're in an offline meeting or on the road while driving and can't open your computer to write a document, "One Bean" becomes your external brain.

In short, what software companies are best at has now become the core competitiveness of hardware.

3. ByteDance, Meta, and OpenAI are all betting on one thing

You may ask, since software is so powerful, why not just make apps as before? Why do they have to get involved in making hardware themselves?

Imagine the scenario in 2026. If you're wearing Meta's glasses or OpenAI's headphones and you have a need. For example, you want to schedule a meeting at 3 p.m.

Will you take out your phone, unlock it, find the Feishu icon, click on it, and then look for the calendar?

Most likely not. You'll just say to the glasses: Schedule a meeting for me.

At this time, the AI Agent in the glasses will directly handle the matter for you. In this process, the Feishu app completely disappears. It becomes a "plug - in" that provides data in the background.

This is fatal for large software companies.

If users no longer open the app, large companies will lose direct control over users. They can't see users' behavior habits, can't push ads to users, and can't sell other value - added services. They become workers for hardware manufacturers.

Feishu's making of "One Bean", Meta's making of glasses, and OpenAI's making of headphones are all essentially about defending their interaction entrances. They must have their own hardware carriers to ensure that when users have a need, they turn to their AI first, not others'.

Whoever controls the hardware closest to users' ears and eyes controls users' habits.

In short, ByteDance, Meta, and OpenAI are all betting on one thing: What if users don't click on apps?

This is a battle for the right to control the user entrance and the first interface.

4. Data scarcity, and even more panic without data

Besides seizing the entrance, there is a more hidden reason: AI has exhausted the data on the Internet.

You may not know that the current AI models are facing an embarrassing reality. The high - quality text data on the Internet is almost used up.

According to a report by the Epoch AI Research Institute, the high - quality English language data may run out between 2026 and 2032. In China, the data of each large company is self - contained and not open.

If AI wants to continue to evolve, it must find new data sources.

Where is this data? It's in our real life.

The data on the Internet is "dead" and processed by others. But the conversations in the real world, the scenes you see, and the tone in your meetings are real - time data.

AI hardware is essentially a 24 - hour detector.

When you wear AI glasses and enter a supermarket or use an AI recording pen in a meeting, AI is learning in real - time how humans handle real - world problems. It can see your logic in choosing products and hear your negotiation skills.

This first - person audio - visual data from the physical world is the scarcest resource for training the next - generation AI at present.

By selling hardware, large software companies are actually installing countless eyes and ears around the world. The data sent back to the cloud can make their models smarter and more understanding of human behavior logic.

Why do large companies make AI hardware? They don't care about the hardware profit. What they want is the real world in front of you.

5. Hardware is no longer a heavy asset

Why do people think they can succeed this time? Because large companies have become smarter and no longer try to handle everything by themselves.

Previously, when large companies made hardware, they took on everything from circuit board design to mold - making and factory production scheduling. As a result, it was not only slow but also lacked professionalism.

Now the trend is cooperation. For example, recently Doubao collaborated with ZTE to make mobile phones, and Feishu's choice to work with Anker Innovations on "One Bean" is a typical sign. It has to be said that ByteDance has mastered the dual - mode of software and hardware well.

Anker Innovations is a world - leading consumer electronics company. Their understanding of hardware such as the supply chain, batteries, and audio chips is world - class, and their supply - chain advantage is very strong. Anker Innovations provides the hardware carrier and distribution channels. Feishu provides the AI model and connects the office ecosystem.

This division of labor has significantly reduced the trial - and - error cost for software companies.

Feishu doesn't need to study how to make a non - heating charging chip, and Anker Innovations doesn't need to train a large - language model. This "Lego - style" assembly model has shortened the hardware R & D cycle from the previous 18 months to the current 6 months or even less.

Hardware is no longer a heavy asset but has become a plug - in that can be quickly iterated.

The production cycle of hardware has been shortened, and the risks have been shared. This makes large companies think that it's worth jumping into this "fire pit" again.

[Beyond the Page] Words:

The AI hardware trend in 2026 seems lively but is actually full of helplessness.

The collective move of software companies into hardware is essentially a fear of the end of the mobile Internet era. People have found that the walls of apps can no longer withstand the impact of AI. If you don't turn yourself into a physical entity that users carry around, you'll be forgotten by users.

But we also need to think about a question: Do we really need so many AI recording pens, glasses, and headphones?

The progress of technology is supposed to make people's lives easier. But now, we seem to be surrounded by more sensors. Large companies are vying for our attention, the scenes in front of us, and the sounds in our ears.

When the large - scale model finds its carrier, it is no longer just an icon in the phone but becomes a part of our body.

Is this human evolution or another deep penetration into our lives?

This article is from the WeChat public account "Beyond the Page", author: Huahua, published by 36Kr with authorization.