Listen to the audio version of this article (generated by artificial intelligence).
The Prisoner’s Dilemma with a Chinese Accent… Another Story in the Chaotic Middle East… OpenAI’s AI Hacks the Hugging Face… Reader Comments on Yesterday’s Report digest
an act alphabet (Google) Confirm capital expenditure?
By the time you read this, we probably already know the answer. Alphabet’s second-quarter earnings report — and the management call that followed it — is due.
As we explained in detail on Monday digestWall Street has become skeptical of the rosy earnings forecasts issued by some AI infrastructure companies this season.
Our technology expert Luke Lango, editor Innovation investorhe was frank about what it will take to change this mood: hyperscalers need to confirm their capex plans.
So, has the alphabet arrived?
We will review its results tomorrow. If capex guidance comes in strong, it’s a green light for AI trading. If not, expect skepticism to deepen and fireworks marketing to intensify.
For now, let’s move on to a different AI story — one that’s less about whether spending holds up, and more about what happens if the very thing that hyperscalers spend on becomes cheaper and more available faster than anyone planned.
Kimi K3, and the swap we said we’d be watching
Back on April 6, and it was addressed yesterday digestwe have set Five prisoner’s dilemmas that run through the AI economy -Moments in which each actor makes a locally rational choice, and yet the sum of those choices produces an outcome that no one actually wants.
The fourth dilemma revolves around the AI infrastructure itself…
Every hyperscaled company must pour billions into computing power or risk being left behind — even as its research teams race to build smarter, more efficient models that need less of the same power.
As we wrote back in April:
The better the AI research team does its job, the more it undermines the infrastructure team’s investments…
So, the company is simultaneously building an empire… and designing the weapon that will destroy it.
Over the past few days, investors have been given one of the clearest real-world examples of this yet.
Two Chinese companies, Moonshot AI and Alibaba, have released open-weight AI models: Kimi K3 and Qwen 3.8 Max. Because it’s open, anyone can download, play, and customize it for free.
They’re said to be excellent – the Kimi K3 beats both Claude Opus 4.8 and GPT-5.5 in programming benchmarks, and at a fraction of the cost.
This is the prisoner’s dilemma that appears outside the walls of any single hyperscalator…
What happens when AI capability continues to improve faster than the economics that underpin today’s infrastructure can accommodate it?
The Wall Street Journal Linking it directly to the recent tensions in the market:
The threat of new players undercutting what they can charge for advanced AI led to a significant drop in the share prices of some technology and AI companies last week.
In other words, the trillion-dollar bet on infrastructure has always been on Anthropic and OpenAI Remaining capable enough to justify the bill.
But a free competitor that is just as good or better not only doesn’t compete, it breaks away from the assumption the full version was priced at.
This is the fourth prisoner’s dilemma, straight away, with a new name attached to it.
But this is not just an example of the prisoner’s dilemma, it is also a story of the “messy middle.”
One title – two frames.
in yesterday digestI introduced a concept related to the prisoner’s dilemma – the chaotic middle.
In short, as our world adapts to AI, we will be forced to navigate competing and clashing priorities. It will not be a matter of choosing the “good choice” over the “bad choice,” but rather it will be two things we desire that cannot be equally chosen at the same time.
We must choose – and live with the consequences of the path we did not choose.
In this case, OpenAI and Anthropic have spent the past week sounding the alarm about a world of cheap, open, and uncensored AI models. Dean Paul, head of strategic futures at OpenAI, doesn’t take kindly to describing this future as a “dystopian hellscape.”
Here’s more from Ball:
One possible outcome of an open, model-dominated world is full-blown AI communism, which is precisely what China is proposing: rather than a product for the market, AI is a “public good” that the state will eventually provide as a kind of “digital public infrastructure.”
Anthropic CEO Dario Amodei has separately spent months warning that freely downloadable models with advanced capabilities pose a real cybersecurity risk — the kind that doesn’t require a rogue superintelligence, just a bad actor with a laptop and no company standing between them and the tool.
We all want and prioritize a safe world. So, we should ban these open source models, right?
Except – what about the other thing we want?
progress.
Cheap and open AI puts high-level tools in the hands of any developer, any small business, anywhere, without subscription fees in the way. The world benefits from lower costs and faster progress.
Isn’t this exactly the democratization of technology we’ve been calling for?
maybe. But the openness that makes free, amazing AI “good” removes every guardrail that a closed, paid model comes with.
Which brings us to a story this morning about what can go wrong…
“An unprecedented cyber incident”
Earlier today, OpenAI revealed that one of its AI models — a next-generation system that has not yet been released publicly — hacked into the infrastructure of Hugging Face, the platform that hosts a large share of the world’s AI models and datasets.
The hack occurred while evaluating the model’s cyber capabilities, where guardrails were intentionally lowered so researchers could see what it could do.
OpenAI hasn’t downplayed the importance of this:
We consider this an unprecedented cyber incident, involving state-of-the-art cyber capabilities.
This is not a warning that AI might one day do this. It is a frontier laboratory that monitors its own model to find a vulnerability, exploit it, and hack third-party systems – under moderated and intentionally mitigated conditions.
However, it gets messier…
The co-founder of Hugging Face did not use this incident to move away from open AI systems, but rather argued the opposite.
His point is that when a frontier model starts attacking your infrastructure, defenders need fast, broad access to tools near their border to respond — not a slow, clunky process of requesting permission from whoever controls the closed model doing the attack.
So, what will we prioritize?
Advances in cutting-edge technology or safety?
This is the messy middle, live in action.
Run it through our lenses
Yesterday digestI said that we will keep going back to two lenses whenever the messy middle appears. Let’s find out what they tell us here.
Lens 1: Which side of the productivity/disruption trade-off is a particular country or company protecting – and is your portfolio exposed to the side that is becoming more difficult?
We don’t have a clean signal yet.
On the one hand, President Trump has called for few, if any, restrictions on artificial intelligence.
On the other hand, today’s OpenAI hack sparks calls for immediate regulation from various politicians like Texas Congressman Greg Casar:
This is very disturbing.
Artificial intelligence is developing so quickly with no real regulations to keep us safe. This must change.
Until Washington chooses a direction, we do not know whether restrictions would favor closed border labs (by weakening their free competition) or whether inaction would favor companies that already save money by operating cheap, open models internally.
This is a real “we don’t know yet” thing, and we’d rather tell you that than force a conclusion that doesn’t exist.
Lens 2: Does this news push Lock Lango’s AI backlash clock to 2028 earlier, or give it more breathing room?
In short, Locke believes the populist backlash against AI will reach a tipping point around the 2028 election cycle, leading to policies and legislation that could derail the AI boom.
Luke watches the rising cost of living and widespread AI-related job losses. Both paths this story could take to fuel that hour, just through different mechanisms.
Allowing free AI models to proliferate removes the two biggest factors slowing the displacement of AI-driven jobs: cost and access.
A company that might refuse to put a price on enterprise AI has no such hesitation about the free price — and no such hesitation about the layoffs that follow. This accelerates the loss of function aspect of Luke’s thesis.
However, restricting open models creates a new complaint: people pay more, or lose access to a free tool, and it seems as if the motive is to enable a handful of US AI companies to protect their margins.
Pay attention to complaints about inequality and systems of injustice that benefit a select few. It’s a cost-of-living complaint, just a different flavor from electricity bills.
Either way, we run the risk of growing anti-AI sentiment that could impact your investment portfolio. There may not be a version of this story that doesn’t have this story in it – which in itself is more interesting than choosing a side.
wrap
Open source Chinese AI models are not just another title. It serves as a reminder that today’s biggest investing questions rarely have clear answers.
AI has the potential to become dramatically cheaper and more powerful… while at the same time creating uncertainty for the very infrastructure companies that make this progress possible. That is the prisoner’s dilemma.
Deciding which outcome we should prioritize – or which investments ultimately pay the most dividends – is a messy middle ground.
We expect to see more of both before this shift is over.
digest Readers weigh in
Yesterday digestI asked for your comments about artificial intelligence and this chaotic environment. We’ve received a lot of feedback. Thanks everyone for writing.
Scott raised J. Good point:
Another “minor risk” of AI deployment… exaggeration, exaggeration and misinformation by the media regarding the many benefits and risks of AI.
For example, trying to suggest that banning all data center development from an entire state is at all the same as banning data center development in residential neighborhoods is a misreading of both cases.
Agreed, Scott.
As for my coverage of both yesterday digestYou’re right that these are not equivalent policies – the outright moratorium and the targeted zoning rule are two different tools with different intentions.
A quick clarification: I was not referring to the identical nature of the policies themselves, but to the same political reaction behind them: officials in both parties responding to domestic resistance.
Meanwhile, Tom, a 77-year-old reader who saw manufacturing fade out of his area decades ago, says he sees the same dynamic playing out again now — people stuck in the gap between the end of an old economic era and a new era that has yet to arrive:
Let’s hope we don’t move forward and become like Eloi (in The Time Machine) but progress can’t be stopped so who knows.
If you didn’t get Tom’s reference, the Eloi are a future human race in HG Wells’ story who have had everything provided for them, and have lost all ability and curiosity as a result. It’s along the same theme as ours Mouse Utopia digest.
We’ll end with Peter C, who wraps up our entire conversation today succinctly:
This is an exciting time to be here because the solutions will be a compromise of some sort. Thanks for the article!
Thank you Peter – and to everyone who wrote.
If you want to participate in the conversation, email me at ipdigestfeedback@investorplace.com.
I wish you a good evening,
Jeff Remsburg
(Disclosure: I own GOOGL)
