There was a time when I was a person who spent without a second thought. I’d casually buy premium groceries from Hema¹ and pay for my parents’ vacations without hesitation. I just flipped through my JD.com² order history, and the change in the numbers is startling:

• 2021: ¥9,791
• 2022: ¥22,206
• 2023: ¥5,778
• 2024: ¥10,000 (for a necessary productivity tool)
• 2025 (to date): ¥1,654 (for staples like rice, flour, and oil)

My five-year-old phone was on its last legs, and I thought about replacing it during this year’s “618” shopping festival³. But faced with the idea of “buying a new phone,” I hesitated for days. I kept asking myself: Would a new one boost my productivity? Would it improve my quality of life? Would keeping the old one negatively impact my daily use? The answer to all three was no. However, the temptation of a 24-month interest-free installment plan and a generous subsidy was just too great. I caved and placed the order. But when the delivery arrived at my door, reason ultimately won out over impulse, and I rejected the package.

In today’s world, where the wave of artificial intelligence is sweeping the globe, GPU computing power is a key factor of production. However, a common pain point is that GPU resources are both scarce and expensive.

Take mainstream cloud providers as an example. Not only are GPU instances often hard to come by, but their prices are also prohibitive. Let’s look at a direct comparison:

• Google Cloud (GCP): The price of one H100 GPU is as high as $11/hour.
• RunPod: The price for equivalent computing power is only $3/hour.
• Hyperstack / Voltage Park: The price is even as low as $1.9/hour.

The price difference is several times over! This leads to a core question:

Can we design a solution that allows us to enjoy the low-cost GPUs from third-party providers while also reusing the mature and elastic infrastructure of cloud providers (such as managed K8s, object storage, load balancers, etc.)?

The answer is yes. This article will detail a hybrid cloud solution based on Tailscale and Kubernetes to cost-effectively build and scale your AI infrastructure.

Recently, I encountered a very strange phenomenon: in a newly created cluster, when doing a subPath mount, the file in the container at that subPath is empty. I checked the syntax and configuration; everything seemed correct. But I’m not hitting the known bug where a non-existent ConfigMap subPath is mounted as an empty file nor issue 54514. Moreover, if I mount the ConfigMap directly, the key’s contents show up fine in the container. It really feels like I’ve run into a bug!

Due to increasingly annoying YouTube ads that Adblock Plus can no longer block, I switched to using uBlock Origin (the most powerful YouTube ad-blocking extension in the world, bar none). However, the Google Chrome Store no longer allows its installation because uBlock Origin is based on Manifest V2, and Chrome is phasing out Manifest V2 extensions.

Moreover, uBlock Origin has no plans to upgrade to Manifest V3. Its alternative extension, uBlock Lite, is based on Manifest V3 but is not as effective or feature-rich as uBlock Origin. If you don’t mind the limitations, you can install it as a replacement.

Google will officially stop supporting Manifest V2 from Chrome version 139 (June 2025). Therefore, I have decided to stick with Chrome 138 indefinitely.

Below, I will share how to install uBlock Origin on Chrome, even though it is no longer available in the Google Chrome Store.

It’s already 2025, and I’m just starting to research AI infrastructure providers. It feels a bit late. My main purpose in this research is to see if there are opportunities to participate in this AI wave.

AI infrastructure services are categorized into providing technical services and solutions, IaaS services (offering GPUs), PaaS, and SaaS services.

Service providers that want to offer technical services and solutions require strong marketing and sales teams. However, those providing IaaS, PaaS, and SaaS products do not need as many marketing and sales resources.

Major cloud providers are already well-established in offering AI-related services, but the emergence of new startups suggests new market opportunities. I am more interested in AI infrastructure startups catering to researchers, small businesses, and developers.

Becoming a technological investor or a user of technology is the easiest way for us to benefit from the AI era.

Recently, I’ve been thinking 🤔. In the current economic landscape, AI is the only growth point, and going with the flow might be easier. So, how can I find my place in this era?

My thought is that AI is a universal technology, just like electricity, computers, and operating systems. Not everyone needs to work in these foundational services, but rather find a position in the ecosystem built on top of them.

Wu Jun once said that, in the future smart era, only about 2% of the population will truly benefit from technological advancements, and these three types of people will enjoy the benefits brought by technology:

1. Those who own core technology
2. Those who invest in core technology
3. Those who apply technology

No tutorials or books needed, master a new language in 1 day (not for 99, not for 9.9, just if you’re willing to use it – classic sales pitch). This isn’t a dream, and it’s not an ad! AI models make learning a new language in no time possible.

Of course, this statement is a bit exaggerated. What I mean is that in the AI era, we no longer need to learn a new language the traditional way—learning first and then practicing. Instead, we can learn directly through practice.

This is my reflection after using Cursor for one month to learn front-end development.