In partnership with
Salutations, Olio aficionados! 👋
Welcome to the 284th edition of Weekly Olio. We’re thrilled to introduce a fresh new twist to your Sundays: Publisher Parmesan, our hand-picked, thoughtfully crafted edition designed to spark inspiration and insights for the week ahead.
It’s the perfect way to unwind, recharge, and prepare for the week with something truly worth savoring.
If you’re new here and you’re looking for more long-form, crispy writing, click the link to subscribe under this GIF 👇
A word from our Sponsors…
Winning Startups Aren't Bigger. They're Leaner.
Founders deploying AI agents across sales, marketing, and customer service are closing more with smaller teams. 65% more sales leads. Less headcount.
Download the free Practical Guide to Agentic GTM for Startups and start building the stack your competitors dream of.
China's AI Strategy Is Becoming Clearer: Don't Win the Race, Change It
For the past few years, the global AI race has been described as a contest between companies with enormous computing power, billions of dollars in capital, and access to the world's most advanced chips.
That narrative naturally favored the United States.
America had Nvidia. It had the hyperscalers. It had OpenAI, Anthropic, Google and Meta. China, facing restrictions on access to advanced American chips, appeared to be fighting the race with one hand tied behind its back.
Then came DeepSeek. Now comes Moonshot AI's Kimi K3.
The significance of Kimi K3 isn't simply that another Chinese model has reached the frontier. It is that Chinese companies are beginning to demonstrate a different way of competing: build models that are good enough to rival the best, make them dramatically cheaper, open them up to developers, and push them into as many markets as possible.
That changes the question investors should be asking. The AI race may not ultimately be won by the company with the smartest model. It may be won by whoever gets the world to use its model.
When Moonshot AI released Kimi K3, the reaction was unusually immediate. Here was a Chinese AI company producing a model that reportedly ranked just behind the latest offerings from OpenAI and Anthropic, despite operating in an environment where access to the world's most advanced computing infrastructure is significantly constrained.
The model itself is impressive. Kimi K3 reportedly has 2.8 trillion parameters, putting it broadly in the range of the largest frontier systems. Its context window can handle nearly 800,000 words in a single prompt, allowing it to work through enormous legal documents, research papers and codebases without losing track of information introduced much earlier in the conversation.
But size isn't what makes K3 particularly interesting. The economics are.
According to Artificial Analysis, Kimi K3 costs roughly $1.95 per task, compared with about $2.75 for Anthropic's leading model at the time of comparison. The difference may look insignificant when you're asking an AI to write an email. Multiply that difference across millions of enterprise tasks and the economics become much harder to ignore.
And then there is the distribution model.
K3 was released as an open-weight model, allowing researchers, developers, startups and governments to download and customize it.
China isn't merely producing another AI model. It is beginning to export an AI ecosystem.
America Built the Frontier. China May Be Trying to Democratize It
The conventional AI playbook has been remarkably consistent. Build the biggest model. Buy the most advanced chips. Construct enormous data centers. Spend billions of dollars. Repeat.
The strategy has worked exceptionally well for American technology companies. Nvidia sits at the heart of the infrastructure stack, while companies such as OpenAI, Anthropic, Google and Meta have access to enormous pools of capital and computing power.
China doesn't have the same freedom.
US export controls have restricted Chinese companies' access to some of the most advanced Nvidia chips. That should, in theory, make it significantly harder for Chinese companies to compete at the frontier.
Yet DeepSeek and Moonshot have demonstrated that there are other ways to close the gap.
The emerging Chinese strategy appears less focused on winning the absolute frontier at any cost and more focused on producing models that are good enough, cheap enough and widely available.
That distinction is critical.
If AI eventually becomes a utility embedded in everything from software development to customer service, the most valuable company may not be the one whose model is 5% smarter.
It may be the one whose model is used by 50% more people.
The Controversial Shortcut: Distillation
This is where the story becomes more complicated.
How are Chinese companies producing such capable models while operating with less access to cutting-edge computing?
One answer being investigated by US officials and AI companies is distillation.
The concept is relatively straightforward. A less capable model learns from the outputs of a more capable model. Rather than developing every capability independently, developers can use the behavior of an existing frontier system as a source of training.
Anthropic has said it identified what it described as industrial-scale attempts by several Chinese AI companies, including Moonshot and DeepSeek, to use its models in this way. The allegation is that large numbers of accounts were used to generate outputs that could then help train competing systems.
The US government is also investigating whether Chinese AI companies obtained restricted American chips in violation of export controls.
Moonshot has not publicly confirmed those allegations, while Chinese officials have emphasized domestic innovation and technological self-reliance.
There is an important distinction here.
Replicating or learning from the capabilities of an existing frontier model is not the same as independently building the next frontier model from scratch.
And that distinction may explain where China currently sits.
It may not yet be leading the race.
But it is finding ways to run surprisingly close behind.
Find the perfect marketing agency, for free.
Stop writing RFPs. Tell us your budget and needs, and our experts send back 3 or 4 agencies worth pitching. The no cost way to source the best agency, without lifting a fingers
China's Strategy May Be to Buy Time
This is perhaps the most important insight for investors.
China may not need to beat America today.
It may simply need to avoid falling too far behind while it builds the rest of the ecosystem.
That means developing domestic chips, expanding computing infrastructure, training local models and, critically, building distribution.
Kimi K3's open-weight approach fits neatly into that strategy.
So do the efforts of other Chinese technology companies to release increasingly capable models at competitive prices. The objective isn't necessarily to prove that China has built the world's most sophisticated AI.
It is to ensure that developers and businesses around the world become comfortable using Chinese AI.
That distinction changes the strategic game.
Imagine a developer in Southeast Asia building an application today. If Kimi, Qwen or DeepSeek performs well enough and costs significantly less than a proprietary American model, there is an economic reason to choose it.
Multiply that decision across millions of developers and businesses over five or ten years and something much larger happens.
An ecosystem emerges.
And once an ecosystem exists, replacing it becomes much harder.
China may therefore be playing a long game: use today's models to build tomorrow's market share, while using that time to develop a fully independent technology stack.
Open AI May Be the Real Battleground
This is why the debate over open-weight AI has become increasingly important in Washington.
At first, the argument appeared straightforward. If advanced AI could be strategically important, governments should restrict access to the most capable systems and hardware.
But there is a counterargument.
If American companies make their models too difficult to access while Chinese companies make theirs widely available, restrictions could inadvertently accelerate the adoption of Chinese technology.
Nvidia CEO Jensen Huang has argued that open-weight models are important for the vibrancy, security and safety of the American AI industry. Meta has also embraced the strategy, releasing its own open-weight models designed to run locally on consumer devices.
This is more than a philosophical debate about open versus closed software.
It is becoming a battle over distribution.
Closed models give companies greater control over monetization, intellectual property and usage.
Open models can spread much faster.
And in a technology race where adoption today can determine ecosystem dominance tomorrow, distribution may ultimately matter more than margins.
Stop Watching Benchmarks. Watch Market Share.
This is where investors should be careful. AI discussions are filled with benchmarks. Parameter counts. Context windows. Reasoning scores. Coding performance.
These numbers are useful, but they can distract from the metric that may matter most.
Usage.
If Kimi K3 is slightly less capable than the best American model but costs substantially less, is open to developers and performs well enough for everyday business tasks, that may be far more commercially significant than winning a benchmark.
The same principle applies to the broader Chinese AI ecosystem.
If American businesses begin deciding that they don't need to pay premium prices for proprietary AI because Chinese models can handle coding, presentations, research and customer service at a fraction of the cost, the consequences could extend far beyond Moonshot.
They would affect OpenAI, Anthropic, Nvidia, cloud providers and the enormous capital expenditure currently being justified by expectations of continually rising AI demand.
The market has spent years asking who has the smartest model.
It may soon need to ask a different question:
Who has the most users?
Closing Thought
The history of technology rarely rewards the company with the most sophisticated product simply because it is the most sophisticated.
The winning product is usually the one that becomes part of the largest ecosystem.
China appears to understand this distinction.
It may not currently have the same access to frontier chips or the same financial resources as the United States. But it can still produce capable models, make them inexpensive, open them to developers and push them aggressively into global markets.
That strategy could look unimpressive in a benchmark comparison.
It could look very different five years from now.
Because if China can use today's AI models to build tomorrow's developer ecosystem, while simultaneously developing its own chips and computing infrastructure, it may eventually stop trying to catch America.
It may simply build a parallel AI world.
And that is the real significance of Kimi K3.
The AI race is no longer just about who can build the smartest machine.
It is becoming a contest over who can make their version of AI impossible for the world to ignore.
Interested in learning more about AI? Check out our previous coverage here:
Reply to everything. Edit nothing.
Your inbox is full. Slack is piling up. Client messages need a response yesterday. Typing thoughtful replies to all of it takes hours you don't have.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Speak like you would to a colleague — tangents and all — and get polished output. Emails, Slack, LinkedIn, WhatsApp, whatever's open.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.
That’s all for this week. If you enjoyed this edition, we’d really appreciate if you shared it with a friend, family member or colleague.
We’ll be back in your inbox 2 PM IST next Sunday. Till then, have a productive week!
Disclaimer: The views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author's employer, organization, committee or other group or individual.