To those who predicted the end of SaaS
Atlassian reported its first quarterly operating profit in more than two years, while Shopify reported 34% quarterly revenue growth. Monday.com grew revenue 22%.
Seriously, you can’t imagine how many people live in the AI bubble, and have the same view as Leopold Aschenbrenner, the 24-year-old kid who bet every dollar that he raised into AI and shorted Saas. A story I talked about in this brief.
The problem with most who think AI is replacing SaaS is that they have no understanding whatsoever of how software works. Yes, it includes that 24-year-old AI scientist genius.
Most only see the feature layers.
But the thing is, companies do not pay SaaS firms just for those shiny features or a colorful dashboard.
They pay for a shared source of truth. Shopify knows what was sold, the transaction records, stock management, customers demographic. Or Atlassian contains years of tickets, approvals, internal knowledge, permissions, and the record of who was meant to do what.
Yes, there are workflows within that software that can be automated.
But here’s my question: what makes you think Atlassian or Shopify won’t and can’t easily automate those themselves by integrating AI, and that a CEO would trust a new vendor founded by a 10-person team, with no track record of delivering stable and risk-free software?
This is different from giving everyone ChatGPT.
Announcing those SaaS redundancies is essentially announcing a reinvention of the entire internal workflow. Maybe it can be done swiftly for a small startup team. But for even a 200-person company? Try to do this w/o droping a few balls.
All that said, it doesn’t mean every SaaS company is safe.
Does that mean all SaaS are safe?
Canva cut its 2026 revenue-growth forecast from roughly 30% to 20%. Figma’s Q3 guide implies 36% growth, a 12-point slowdown, and investors were not thrilled.
→ The Information on Canva, Figma
Unlike other SaaS firms, Canva and Figma face a more immediate AI problem because the majority of their value lies in producing visual output.
Until recently, you needed to learn how to use a tool as simple as Canva, choose a template, and drag things around to make a simple image.
Yet, from late 2025 onwards, you can type what you want into ChatGPT and get a usable output within a few seconds. And generally, you can expect a higher-quality outcome if you provide a reference point and a good enough prompt.
Generally speaking, there is much less operational risk in replacing Canva with GPT-image-2 for a marketing team than replacing Shopify for a company’s entire commerce operation.
And Canva has a second problem: it has added AI features itself.
Just like all other old-timer SaaS products, it knows it’s inevitable that it will integrate AI into its offering. However, unlike other SaaS products, the extent to which Canva can be automated is limited. And the biggest win here is AI creating an image directly.
That puts Canva in a super awkward position.
Similarly, Figma.
If we are still in 2024, then no AI can replace the collaboration, design systems, version control, and handoff from designers to developers.
But all these obstacles are now gone.
AI can compress a lot of the work that gets people into Figma in the first place.
I won’t say this is the reason why investors are nervous, again, only a handful of investors understand why Figma and Monday would be impacted differently. But their recent market moves are more like a shark smells blood.
Before we continue, are you looking for a thorough analysis of a SaaS company or want to understand a macro trend affecting companies you care about? Book a discovery session here.
OpenAI’s iPhone moment?
OpenAI’s first hardware device is designed to make AI easier to use (as Altman claimed). Each may cost $300, have a doughnut-like shape, and be roughly the size of a hockey puck.
As I mentioned in an earlier brief, AI companies are flocking to hardware because the model itself may become the least defensible part of the stack. Just think Meta and Ray-Ban glasses, or Midjourney and its full-body medical ultrasound scanning chamber.
OpenAI is making the same bet.
It wants to own the object (or see it as a distribution channel) that people use to access AI, rather than remain an invisible model that can be routed away at any time.
The next big thing in AI (after memory)
Sony and TSMC will invest $4.69 billion in a Kumamoto joint venture to develop and manufacture next-generation image sensors, with volume production expected in 2029.
→ Reuters
Everyone is obsessed with AI memory, as I mentioned in my Micron analysis:
But AI also needs eyes if the ultimate goal is to step into our physical world. An image sensor converts light into data. Without it, an AI system has no idea what is in front of it.
This is a strategic move into robotics. In plain English:
AI model = brain
Image sensor = eye
Robot/car/server = body or system
An image sensor converts light into digital information, and AI then uses that information to recognize objects or traffic signs, for example, to calculate its next move.
The next AI bottleneck may be perception.
A model is useless in a robot if it cannot reliably see what’s happening.
That starts with the sensor, so the Sony and TSMC’s strategic move.
A future sensor has to solve a problem before AI gets involved: it needs to give the AI a clean picture to work with. For example, in dim light, a tiny phone camera simply does not catch enough light, so the image gets grainy; or in fast video, the sensor also has to push a huge amount of image data off the chip quickly enough that moving faces and cars do not smear or bend.
Given that the AI’s output can only be as good as its input, in this case, what the camera sees.
So the pressure is behind Sony and TSMC’s work in Kumamoto for more advanced image sensors, aimed at capturing better raw images with adequate hardware efficiency.
An upper limit on how much is spent on AI?
NVIDIA announced a partnership with major financial institutions, including Apollo Global Management, BlackRock, Blackstone, and others, to establish independent compute-financing platforms targeting over $500 billion in third-party capital.
Don’t think of this $500 billion as a funding announcement, but a test of the entire AI economy.
A test of whether people will use enough of AI, for long enough, to pay for everything being built.
Simply put, this $500B is Nvidia helping investors fund the companies that buy its machines and build AI data centers. AI companies then rent those machines. Their revenue is supposed to repay the loans and reward the investors.
Along this journey, Huang is asking Wall Street to view Nvidia’s systems as infrastructure that can consistently generate income, rather than as expensive electronics that become obsolete every few years.
Beyond securing funding to house AI, Haung isn’t just sitting around and waiting for his tenants to arrive. He also gives credits and loans to his customers. To complete the cycle.
So far, the entire AI investment seems like a never-ending uphill climb with minimal downside for AI players.
That said, someone has to bear the risks.
The lenders are left waiting to see whether most companies will keep paying to use AI. If AI demand disappoints, the pain lands with whoever financed and runs the facilities.
I’m not here (at least not in this piece) arguing that the AI bubble is about to pop.
Because some of that demand is obviously real.
But financing can bring capacity online before the final customers find AI useful enough to keep funding it.
Until companies outside the AI industry generate enough revenue to cover the bill, $500 billion doesn’t tell us whether the money will earn its way back.
The increasing Capex bet is worldwide
Tencent Holdings said its Capex in the second quarter nearly tripled from a year earlier to 52.8 billion yuan ($7.8 billion) to train better AI models and meet growing demand for its coding and productivity tools.
→ WSJ
Tencent spent almost three times more than last year on chips, servers, and computing infrastructure for AI.
Why?
Tencent says it needs more computing power to train its models and support growing use of its coding and productivity tools.
I want to share this with you because I find it interesting to see that AI spending is not just an American phenomenon. Read the latest Google and Microsoft earnings analysis.
Even with a distinct software ecosystem, Chinese companies are also investing heavily in infrastructure with pretty much the same reasons.
That said, the spending still only proves investment, not profit.
Tencent is betting that people will keep using its AI tools enough to justify the cost.
Fun facts
Here I am to bust my own bubble (and maybe some of yours, too)!
only 6% of US adults use Claude. It’s funny how people in tech, like me, often think everyone on Earth should now know and often switch between Claude and Codex, but the evidence says otherwise.
29 people out of 100 think AI equals chatbot.
Men are more likely to see the positive in AI, while women are more concerned about AI advancing too quickly or other potential risks.
