Today, we're talking about bot traffic, Chinese robots, AI economics, and the matter of slowing down recursive self-improvement. #
Wednesday. Today’s the day! Microsoft, Meta, Qualcomm, Robinhood and ARM will report results later this afternoon. SoFi already dropped this morning, and its investors are not stoked.
Investors are already unhappy with Alphabet’s quarterly report (our coverage here), so the market will have its eyes peeled on Microsoft and Meta’s statements around capex, AI revenues and their compute situation.
Meanwhile, Iran and the United States continue to mire themselves in a conflict with no easy off-ramps. Recent Iranian strikes are expected to engender a major response from the United States. Commodity prices aren’t reacting well to the news.
Today, we’re talking about bot traffic, Chinese robots, AI economics, and the matter of slowing down recursive self-improvement. To work! — Alex
📈 Trending Up #
Public corruption…bad governance…sour grapes…a boom in the trades… venture funding forAgon,Nuclear Turbines,German defense companies,Freehand…prudism in speech-land
Bot detection is big business: Why is the Internet such a pain to use lately? Not only do we have to wade through GDPR pop-ups, many websites now also want to make sure we’re human. Nothing like waiting for Cloudflare to lift the rope and let you into the club!
The problem is that bot traffic is swamping websites, agents are running amok trying to scrape everything in sight, digital advertising remains an oligarchy, and search no longer directs as much human traffic as it once did (for many websites, at least).
So! What to do? Get better at telling biological sources of clicks from digital ones. That’s what Spur Intelligence does, and it claims that in the second quarter, it saw 210% “year-over-year new business growth,” 109% NDR, 0% gross churn, and “42% year-over-year growth in annual recurring revenue.”
Spur was founded in 2017, so those are pretty bangin’ numbers for a nearly decade-old company. They’re in fact so good that Insight Partners just led a $200 million round in the company.
Spur does more than bot detection, but I think we’re going to need a two-lane Internet: one for bots, with more API hooks than GUIs, and another for people, with a greater focus on the interface. Otherwise, we’re going to find ourselves constantly stuck in line waiting for the bots to get a clean bill of health.
- The x402 crew are cooking upanother approach: Why not merely charge bots a tiny fee for accessing websites? That would solve almost all of our current problems, provided the bots abide by the rules. Many won’t. - I suppose that’s why Insight just invested in Spur.
Protectionism: All that shouting spurred by reports that Washington may try to restrict the use of Chinese open-weight AI models Stateside was partially grounded in the need to preserve internecine AI lab fisticuffs:
Open-weight letter signed by 100+ U.S. tech companies: “Open weights also strengthen competition and competition is what keeps the gains of AI broadly shared rather than concentrated in a few hands.”Zuck’s recent op-ed, arguing for a no-holds-barred AI market: “If only a handful of institutions have superintelligence, they will inevitably exercise a controlling influence over economics, science and politics. Even with the best intentions, that concentration would limit people’s ability to choose their own future.”
Message received: more competition is more good.
Enter the FCC, which has added Chinese robots (both humanoid and quadrupeds) to its covered list alongside certain electricity-generating technology (the precise impact the Chinese inverter ban will have on domestic solar is interesting).
What now? The regulator frames the robot decision as a security move: The networked capabilities of advanced robotic systems create extensive vulnerabilities and vectors for attacks that can manipulate the data and physical operation of the advanced robotic system. Relying on foreign-produced advanced robotic devices presents unacceptable supply chain and cybersecurity vulnerabilities […] Advanced robotic devices collect data that could be leveraged by malign actors to surveil Americans, enhance the capabilities of foreign intelligence services, or to remotely commandeer the robots.
No matter how much stake you put behind the FCC’s logic, the result of its decision will be lower competition for domestic AI robotics companies. Great news if you own Figure shares, say, or are eyeing the Agility Robotics SPAC; Bad news if you simply wanted faster and cheaper robots sooner.
- One can make
[security claims about Chinese AI models](https://www.semafor.com/article/07/29/2026/censorship-in-chinese-ai-models-can-be-undone-new-research-shows), too, for what it’s worth.
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
Shame…keeping staff put…our collective cybersecurity…car-based deliveries…Anthropic’s rep…OpenRouter’s effective revenue multiple as it reaches $140M in annualized top line
**The data center gear trade: **You’d think investors would be stoked to see an industry giant’s revenue soar 257%. Not if you’re SK Hynix’s backers, apparently, who are currently punishing the company for reporting that revenue increase, a 557% surge in operating profit, and EBITDA growth of 411%. Down go shares of SK, after its $54.55 billion quarterly revenue came in below expectations.
Other recent high-flyers in the “sell compute gear to data centers” industry, like Sandisk (SSDs, massive growth, new earnings in early August) are also shedding value.
Is it profit taking? A reversal of momentum? What goes up must come down? Ask your local crystal ball. But I think we can safely assume that some of the most recently bounced tech stocks are finding their way back towards Earth once again.
- We of the Index Fund Gang are just riding out the ups and downs; if you are tradingthis excitable stock market, hats off. You have thicker stomach lining than we here at.CO
Beef in the HRtech space: Remember the Rippling-Deel fracas? Rippling (payroll tech and other services) had *alleged *that Deel (focused on helping companies hire remote staff) hired a mole to spy on its work and steal secrets.
The claims seemed pretty cut and dry, especially since an embarrassing honeypot appeared to catch Deel red-handed. The case is ongoing, and Deel is firing back.
Now, Runlayer is suing Rippling, after “an extensive product trial conducted by Rippling as a prospective customer, during which the MCP startup shared everything from its product roadmap to its actual source code,” TechCrunch reports.
The two companies, per TC, could not agree on price, and the deal fell apart. Then a source told Runlayer’s CEO that Rippling was allegedly cloning the product it had declined to purchase.
Oof. Rippling confirmed it is building a competing MCP gateway (more here), but didn’t ascribe much weight to the MCP startup’s claims, saying, “Runlayer’s panicked effort to avoid competition by fabricating claims is not an effective way to deal with its business failures.”
Rippling could have walked away from Runlayer after finding its product to be sub-standard, and decided to build its own. Totally fair. If it used the detail-sharing period it enjoyed with Runlayer to extract critical technology, well, that’s a different story. I suspect we’ll learn a lot more if the lawsuit advances.
I didn’t expect the somewhat buttoned-up HRtech market to be such a hotbed of legal action. It’d be great if some of these major players could go public so that we could clear some of the fog surrounding their results and operations.
Wouldn’t a bit more transparency be great? I think so!
We now understand how Chinese AI labs will generate sufficient revenue
How much do you care about inference margins? That much? Excellent, let’s talk a little theory.
Neoclouds love open models, especially those that sit near the frontier. Because historically, AI compute providers could load new models and serve inference to anyone who came calling.
OpenRouter lists nearly two dozen inference providers for Moonshot’s Kimi K2.6 model, for example, and they’re all competing for inference market share. It’s discounts galore, you can even choose providers from different continents.
Kimi K3, Moonshot’s latest model — and the inheritor of DeepSeek R1’s mantle of making the American government change its pants — by contrast has a mere eight inference providers listed, one of which is the AI lab itself.
Why? Moonshot wants a cut. Here’s the critical portion of Kimi K3’s terms (emphasis added): 2.
“Model as a Service” means giving a third party access to language model inference or fine-tuning (e.g., via API) in a manner that allows such third party to exercise meaningful control over the inputs, parameters, or training data.This does not include (a) end-user products with model capabilities solely
embedded within specific features or harnesses, or (b) mere relaying of requests to models hosted by others.
If the Licensee or any of its affiliates operates a Model as a Service business, and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months,the Licensee must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose. In other words, if you want to sling Kimi K3 to your neocloud customers, you have to sign papers with Moonshot. That means the lab gets a cut.
This creates a new dynamic:
-
Third-party inference providers for Kimi K3 have to pay Moonshot for the privilege.
-
Presumably, this may apply to subsequent models from Chinese AI labs that want to grow more quickly, which is all of them.
-
As a result, third-party inference providers will operate at a disadvantage to Moonshot’s own pricing, leaving the AI lab with a structural price advantage.
-
Moonshot is shifting neolabs’ margins from their books to its own income statement.
-
This will create more revenue for Moonshot, which can then plow the funds back into creating new and better models.
Is this a deathblow to the neocloud model? Not at all. But it is some sort of evidence that there’s a high-margin business in serving compute for someone else’s model
On the other hand: Can one consider Moonshot’s move as a one-off rather than the start of a new relationship between Western compute providers and Chinese AI labs?
The AI lab just raised $3.5 billion in a round that Bloomberg reports was “larger-than-anticipated,” at $35 billion valuation. No one invests that much money without wanting fat returns.
If Moonshot simply tried to compete on vanilla pricing with neoclouds, it would capture too little of the total value its models create. Thus, over time, we should expect to see more commercial terms between the companies that want to serve open-weight Chinese AI models and the models’ developers.
- The terms do not preclude companies from using K3 internally, or tuning it. The moment you do that for others, however, you gotta shell out.
Is early recursive self-improvement the right time for an AI slowdown?
No.
Subscribe to Cautious Optimism to unlock the rest.
Become a paying subscriber of Cautious Optimism to get access to this post and other subscriber-only content.