Welcome to the Halftime Show.

It’s CES week. 95% of the people I know in Shenzhen are in Las Vegas shoving buffet down their throats right now. The rest are watching the livestreams and taking notes. Come with me next year?

Let’s lock in.

CES is the super bowl of tech - where the world comes to see what’s next. In sin city Las Vegas. 148,000 attendees. 4,100 exhibitors. I couldn’t make it this year, and the FOMO is torturous.

Everyone came for the robots doing backflips and the AI pet that gets lonely. And the hardware looked genuinely epic. One MIT reporter put it best: the confidence wasn't about any single breakthrough. It was "we can iterate faster than anyone else."

I think that’s actually an infrastructure statement. And it changes everything about what CES 2026 actually meant. And what the years beyond will look like.

→ Jensen Huang shared his stage with 4 Chinese humanoid robots. First post COVID CES where China's presence was genuinely unmissable. From the floor, to whale meetings, to engineering demos. Veteran attendees said it felt different this year. (MIT Technology Review)

→ Four robots. Four companies. Every room of your life covered. Unitree for the factory. Fourier for elderly care. Zhipu Robotics for the living room. Woan for the home. B2B and B2C. Everyone gets one.(CES 2026 exhibitor data)

→ 18 companies showed up to CES selling robots that get lonely, hungry, and go through a rebellious phase. Companion robots drew the biggest crowds after humanoids. Desktop pets with body temperature. Pocket devices with memory. All Chinese startups. All heading to mass market pricing within two years. (36Kr)

→ Geely navigated an underground carpark without a driver. On stage at CES. L3 capable autonomous driving across 16 production models. An automaker most people outside China have never heard of. Shipping now. (China Daily)

→ Alibaba released a Google Gemini rival. Free. Open source. Self hostable. Qwen-Image-2512. Apache 2.0. Deploy it yourself, modify it, commercialise it. No vendor lock in. No API dependency and no cost. (Open Source For You)

Lenovo's main stage event wasn't about laptops - but a cross device AI agent system and a Nvidia partnership. So the real story was in the back half of the booths. Frameworks. Tooling. Enterprise deployments. Hybrid AI. Cloud platforms. This is laying the sticky groundwork for a dynasty to come.

The companies shipping the cute robots are building the enterprise infrastructure layer at the same time. That combination- consumer spectacle on top, industrial stack underneath - is the compounding delta that you need to pay attention to.

Take SenseTime. Most people clocked their chess robot on the show floor. Fewer know their [SenseNova V6] model is already processing insurance claims, video analysis, and enterprise workflows across multiple industries (600 billion parameters, multimodal reasoning ranked first in China against GPT-o1). Same company. Consumer spectacle on the show floor. Enterprise AI infrastructure underneath it.

And there’s hundreds more like them.

Every week of iteration, every open source contribution, every developer building on top of Qwen or DeepSeek widens the moat. Not because any single week matters. Because fifty two weeks of that, every year, does.

For founders and operators: the question isn't whether these models match GPT or Claude on a benchmark. It's whether they're good enough for your use case - at a price point that changes your unit economics. Increasingly, the answer is yes.