Hey👋,
I'm Giacomo

Zero-click searches are declining.

March 2026 saw the lowest share of Google desktop zero-click searches in the last two years, according to Datos, A Semrush Company and Rand Fishkin.

It doesn't mean the trend is completely reversing (still 1 in 5 searches end with zero clicks), but it's a signal.

After the messy early days of AI Overviews, Google is now trying to solve an almost impossible problem:

make AI Overviews useful for users,
without damaging its relationship with publishers.

To do so, it made the following changes over the past few months:

- More sources, more prominent.
AI Overviews now surface more links, prominently called-out in separate panels or increasingly in-line within the answer.

- Different query mix.
AI Overviews used to be mostly informational. That is changing.
According to Semrush, between October 2024 and October 2025, queries triggering AI Overviews shifted:
• Commercial: 8% → 18.5%
• Transactional: 2% → 14%
• Navigational: 0.8% → 10%
Those type of searches naturally need a click.

- Difference source mix.
A 2026 study by the Washington University in St. Louis found that almost 30% of cited domains in AIO did not appear on the traditional first page of results. Social, videos and forum discussions now represent a significant share of AIO sources and links, but they rarely rank on the traditional SERP.

- More focus on commerce.
Amazon and eBay climbed the ranking of the most cited sources by LLMs.
Their links naturally drive more clicks.

What does this mean for marketing?
I wish I knew 😅.

But what's clear is that our job is becoming harder.
Now we need to do three things at once:

1. Rank in traditional search.
Very important as AI search still represents a small share of usage.

2. Be selected as a source in AI answers.
Different from number 1.

3. Be present where discussions take place.
LLMs increasingly value what real users have to say about a product or service.

It'll be interesting to see how the search landscape will look like in a year from now.

SpaceX says its total addressable market is around $28.5tn.

That number is already absurd.
For comparison, the US GDP in 2025 was around $31tn.

But the breakdown is even more interesting.

According to the S-1 filings, SpaceX sees three main markets:

1. AI: $26.5tn
2. Connectivity: $1.6tn
3. Space: $370bn

SpaceX’s future will not be about rockets.
It will be about AI. What a surprise!

But I’d be careful about dismissing this as just “AI hype”.

The core IPO story is the production and launch of orbital AI data centres.

And this is not as crazy as it sounds.
Several companies are already working on it, and the main bottleneck seems to be launch cost.

This is where SpaceX comes in.

When Starship enters service (and it’s not an if, it’s just a when) launch costs could drop to as low as $185 per kilo.
That is low enough to make the economics of orbital data centres work.

Data centres in space can be an order of magnitude more efficient and powerful than those on Earth. Plus, unlike on Earth, you can build as many as you want.
The sky is literally the limit.

These data centres could power the next generation of AI models.
For xAI, yes.
But more importantly, for third-party models as well.
Anthropic is already renting SpaceX’s Earth-based data centres for $15bn a year.

This is not a speculative market.
It already exists.

And if you pair extreme computing power with Tesla’s autonomous driving and Optimus robots, you start to understand the full scope of Elon’s ambitions.

Another extremely interesting segment is Starlink Mobile and its potential enterprise applications.

Think about what high-speed connectivity in the middle of the Pacific Ocean could enable: floating data centres, underwater factories, military applications, and god knows what else.

Starlink Mobile’s consumer applications are also massive.

The S-1 filings estimate that 5.5 billion people will have a satellite mobile data subscription in the near future.

That would not surprise me.
I’m sick and tired of scummy mobile plan conditions and high prices, only to be left with no connection when I need it most.

Starlink will make mobile internet truly universal.
For anyone, anywhere, at a fair price. And I can't wait!

Forget about Mars and asteroids.
The opportunities I mentioned are already enough to justify a huge potential market.

In fact, the Financial Times estimates that SpaceX’s target valuation is justified if the company can capture just a 3% slice of the TAM mentioned above.

Well, right now, it is close to 100% of it.
We’ll see what competitors do!

Sources: SpaceX S-1 filings, Financial Times, The Economist
Infographic courtesy of ChatGPT.

This is the second instalment in a SpaceX IPO mini series.
I published the first post in the series yesterday.
Tomorrow you'll find the third and last one, focused on the most speculative bets.

As always, the Swiss train bringing me to Italy was fairly on time.
It arrived in Milan with a 15-minute delay. I had to run a bit to catch my next train, but that was no problem.

The next Italian train left right on the minute. It arrived at my destination 5 minutes early.

Enough time for me to jump back on the train to grab the suitcase I had left in the luggage rack 😂.

Unfortunately, not enough time to jump back off with it 😅.
So I had to stay on and get off at the next station.

20 minutes later, I caught another train back. This one was also perfectly on time.

On the way back to Switzerland, I was warned about a large strike causing massive disruptions to buses and trains.

My bus to the train station was perfectly on time.
My train to Milan left with a 5-minute delay and arrived right on the minute.

Turns out, the strike was actually limited to a few selected minor routes.
But of course, it was enough for people to claim that “in Italy, nothing works.”

A lot of negative things are said about Italy. Many of them are false or misinformed.

In the picture: the new shiny EuroCity Swiss train from Zurich to Milan.

How would you feel if your colleague upstairs got a $400k bonus while you received just $4,000?

That is what just happened at Samsung.

And it reveals an aspect of the AI frenzy I had honestly overlooked until today.

Until recently, the AI computing story was all about GPUs, mostly designed by NVIDIA.
We all know this story.

But lately, attention has shifted to memory chips, specifically DRAM.
In simple terms, the 16GB, 32GB, 48GB of your laptop.

Only recently did it become clear there is a shortage of advanced memory chips.

LLMs require increasing amounts of memory to operate, as context windows become larger and models more sophisticated.

Even the most advanced GPU doesn’t work efficiently without enough memory.

The result is a massive surge in demand for memory chips.

And an explosion in memory chip stocks!

• SK hynix stock is up 231% year-to-date
• Micron Technology is up 194%
• Samsung Electronics is up 139%

Beware, these sky-high valuations are not just hype. Profits exploded too.

Q1 2026 net income growth vs. previous year:

• SK hynix: +398%
• Micron: +180%

And Samsung: +476%!!

Samsung obviously makes far more than just memory chips:

Smartphones, TVs etc.

The irony is that the consumer electronics business unit is now struggling because of higher chip prices, which in turn massively benefited the chip divisions.

Unions representing Samsung’s chip business units negotiated an unprecedented profit-sharing scheme worth roughly $400k per employee on average.

At the same time, unions from the consumer electronics division requested a court injunction against the scheme.
Their bonus was mostly untouched 😔.

The AI revolution is creating very strange winners and losers.

Love this quote. It's meant for AI investment, but it applies to everything in life. To a career, relationships, friendships.

The risk of not risking is far greater than the risk of risking 🙂.

Everyone wants to become “strategic”.
They want to leave execution to others.
That is what promotions are for, they think.

But when they do get promoted and become “strategic”, they often lose touch with reality.

They stop understanding how the actual work is done.
They stop noticing quality problems early.
They stop learning.

I was reading Ron Friedman, Ph.D. HBR article on "superteams", and one finding stood out.

After surveying more than 6,000 knowledge workers, Friedman found what superteams do differently:

Superteam leaders stay close to the work.
They roll up their sleeves and contribute alongside their team.

The chart below speaks for itself.
78% of superteam employees said their managers were actively involved in the work, vs. 55% for average teams.

They don't stay close to the work to micromanage or to control every detail, but to understand what is really happening. And most importantly, to keep learning and improving.

It might sound obvious.
But in many corporate cultures, seniority equals distance from execution.
The leader "made it".
Now others do the work.

This culture may have been dangerous before.
In the age of AI, it is disastrous.

Technology is moving so fast that no leader can afford to stay hands-off.

AI makes hands-on understanding more important than ever, because the baseline keeps rising. What used to be complex is becoming easier. So to compete, teams need to keep raising the bar.

Superteams experiment, make mistakes, stay curious, learn from one another and practice continuous improvement. This is mandatory in the age of AI.

But that culture cannot exist if leaders talk about continuous improvement from a distance without practicing it themselves.

HBR article: https://lnkd.in/eKyf8K9U

It took Meta 15 years to build a $200B ad business.

‍

OpenAI plans to get halfway there in 4. 🤯

‍

Pinterest has been selling ads for 9 years.

They made $4.2B last year.

‍

Snap has been at it for 10 years.

~$5.2B in 2025.

‍

OpenAI plans to reach half of Meta's size by 2030.

Starting from basically zero!

‍

Here's their roadmap, leaked to investors via Axios reporting:

2026: $2.5B

2027: $11B

2028: $25B

2029: $53B

2030: $100B

‍

That's a growth curve with no precedent in the history of online advertising.

OpenAI is forecasting a 152% ad revenue CAGR over four years.

More than double TikTok's, the fastest-growing ad platform in history so far, which grew at an estimated 71% CAGR between 2021 and 2025.

‍

But it gets even more interesting.

I did the math on revenue per user.

‍

ChatGPT reported 900 million weekly active users as of February this year.

Ads only run on the Free and Go tiers, roughly 95% of users, ~850M.

That means OpenAI would earn around $3 per user in 2026.

‍

They target 2.75B weekly users by 2030, so ~$40 per user.

For reference,Meta currently earns ~$59 per user.

So OpenAI's 2030 target requires 68% of Meta's current monetisation rate.

‍

Ambitious is an understatement.

But not impossible, given the breakneck user growth ChatGPT has already experienced.

‍

Plus, OpenAI ads also offer something other platforms don't.

The depth of data it collects about its users is unprecedented, and extremely valuable to advertisers.

‍

Imagine your personal assistant and mentor selling your conversations to the highest bidder. That's essentially what's happening.

‍

But the real question isn't whether $100B is achievable.

‍

It's what happens to the rest of the ad industry when 2.75 billion people are asking an AI instead of scrolling a feed or typing a search query.

‍

Sources: Axios, Reuters, TechCrunch and full-year earnings report.

Charts courtesy of Claude.

Interesting to see LinkedIn climbing the ranking.

I noticed it first-hand on Google AI Overviews.

‍

It often cites professional and insightful posts, even those with very little engagement. I saw posts with just one like from small accounts being cited as the top source.

‍

It seems LLMs get it.It’s not all about virality. Sometimes content quality matters too.

Source: Semrush via Andreessen Horowitz Instagram account

Most ChatGPT vs Claude takes are wrong.

‍

They ignore the one thing that actually matters.

‍

It’s not ChatGPT vs Claude.

It’s quota vs quota!

‍

Opus 4.6 Extended Thinking > GPT 5.4 Extended Thinking.

But to make Opus 4.6 Extended useful, you need Claude Max.

There’s no way around it.

‍

On Claude Pro ($20/month), the quota is too limited.

For serious work, it makes Opus 4.6 almost unusable.

Claude Max starts at $100/month!

‍

ChatGPT Pro 5.4 Extended > Opus 4.6 Extended Thinking.

‍

But again,

ChatGPT Pro is $200/month.😅

And GPT Pro 5.4 Extended is available only there.

‍

That’s why this is not a model comparison.

It’s a budget comparison.💰

‍

If you have the budget, ChatGPT Pro is the best option for most tasks, including research, coding, images and videos.

‍

If you compare $20/month plans like for like, ChatGPT still wins.

‍

Not because it has the best model.

But because the quota for complex queries is higher.

‍

And in real workflows, that matters more than people think.

I keep hitting the quota on Claude, I almost never hit it in ChatGPT.

‍

Few extra notes:

• Claude is better with slides, and pptx files.

• The Claude add-in for PowerPoint and Excel is genuinely excellent.

• Claude Artifacts can build local interactive dashboards, which is cool, but for me it’s still a nice-to-have rather than something I’d use in production.

• ChatGPT wins in marketing because it also handles images and video, and apps.

‍

For vibe-coding, Claude Code and OpenAI Codex are broadly comparable.

But even there, the same issue comes back.

‍

If you want the full power of Claude Code, you’ll likely choose Opus 4.6 Extended. And then you hit the quota wall all over again.

Extremely frustrating actually.

‍

My conclusion:

For most people, the real decision is not Claude vs. ChatGPT.

‍

It’s how much you’re willing to spend on AI every month.

‍

That’s it.

Tried Starlink for the first time today.

‍

On an airBaltic flight, apparently the first (and only?) European airline offering it. And it’s free.

‍

In fact, I’m writing this post mid-air!

‍

It’s faster than any complimentary WiFi I’ve ever tried, it feels like normal home internet.

‍

A real game changer.

‍

Looking forward to the Starlink IPO. I almost never buy individual stocks, but I might make an exception here.

‍

Cheers from the Baltics!