Tomasz TunguzModify

2025-08-01 08:00:00

What force could dethrone AWS after more than a decade of unchallenged dominance?

For years, Amazon Web Services ruled the cloud infrastructure market. It was the default choice without a question for every startup.

Then OpenAI released GPT-4. Microsoft’s exclusive partnership with OpenAI transformed Azure from a second-place player into the obvious choice for AI-first companies. With this week’s earnings, we are seeing the ultimate impact of that strategic decision.

The numbers reveal a market in transition. AWS generates $30.6B in quarterly revenue compared to Azure’s $22.9B and Google Cloud’s $12.5B, but absolute size masks the real story of momentum shifting beneath the surface.

Since GPT-4’s launch, Azure has consistently added more to its ARR than AWS. In two of the previous eight quarters, Google has booked more new ARR than Amazon.

Jamin Ball’s data highlights the trend. Azure surged from 35.8% market share in Q1 2022 to 46.5% during the GPT-4 launch in Q2 2023, seizing first place through its OpenAI advantage.

Google Cloud has captured 6.4 percentage points of market share since Q1 2022, growing from 19.1% to 25.5% in Q2 2025.

Both Microsoft & Google have stronger AI value propositions than Amazon with OpenAI models & Gemini models. And it shows in their growth rates: Microsoft’s and Google’s growth rates now exceed 39% and 32%, respectively, and are accelerating. Meanwhile, Amazon’s growth rate is flat at 17%.

The market explosion tells an even more dramatic story. Total quarterly ARR additions grew from $5.9B in Q1 2022 to $21.4B in Q2 2025—a four-fold increase that reflects AI’s transformative impact on enterprise spending.

Put another way, Google’s new ARR bookings in the last quarter is the size of the whole industry’s bookings just three years ago.

With Azure and Google Cloud Platform growing faster than AWS, the once-strong incumbent’s market position may lead to three equal players. The next trillion dollars in cloud revenue will flow to the platforms that best integrate AI into every layer of their stack.

2025-07-28 08:00:00

Why is the sub-$5 million seed round shrinking?

A decade ago, these smaller rounds formed the backbone of startup financing, comprising over 70% of all seed deals. Today, PitchBook data reveals that figure has plummeted to less than half.

The numbers tell a stark story. Sub-$5M deals declined from 62.5% in 2015 to 33% in 2024. This 29.5 percentage point drop fundamentally reshaped how startups raise their first institutional capital.

Three forces drove this transformation. We can decompose the decline to understand what reduced the small seed round & why it matters for founders today.

VC fundraising dynamics represent the largest driver, accounting for 46% of the decline. US venture capital fundraising nearly doubled from $42.3B in 2015 to $81.2B in 2024. The correlation of -0.68 between sub-$5M deals & VC fundraising shows a powerful relationship: as funds grew larger, small rounds became scarcer.

Larger funds need larger checks to move the needle. A $500M fund can’t build a portfolio writing $1M checks. The math simply doesn’t work for their economics.

Inflation represents the smallest contributor at just 15% of the decline. What cost $5M in 2015 requires $6.7M today. This represents a meaningful increase but not the primary culprit.

Crucially, BLS data shows software engineering salaries grew at nearly the same rate as general inflation. This means the real cost of building startups remained relatively stable. Salary inflation isn’t driving founders to raise larger rounds.

The remaining 39% stems from other market forces. These likely include heightened competition for deals, increased pre-seed valuations pushing up seed round sizes, & founders’ growing capital appetites as they chase more ambitious visions from day one. The proliferation of seed funds & the emergence of multi-stage firms investing earlier also contribute to this shift.

Here’s the paradox : despite these larger rounds, startups are actually shrinking. Carta data shows SaaS companies are 20% smaller at Series A today than in H1 2020. Smaller teams are more than offsetting inflation costs through increased productivity.

This efficiency gain will accelerate with AI. As productivity tools enable founders to build more with less, we’ll see teams generate more ARR per employee while valuations continue to climb. The best founders are already achieving with five engineers what previously required twenty.

We’re witnessing a shift in startup financing. The small, disciplined seed round that launched thousands of companies in the past decade has been replaced by bigger rounds, higher valuations, compressed timelines, & loftier expansion expectations.

2025-07-23 08:00:00

The internet is about to look a whole lot more like the online advertising world.

No, I don’t mean there’ll be more ads. In fact, I think there’ll be far fewer.

But the technology stack for content distribution will mirror the architecture that has been implemented in the online ad world.

As we’ve reached the AI search tipping point, publishers face an existential challenge : ensuring AI systems use their content in answers to maintain relevancy.

When you visit a website, your browser triggers an auction. The site’s supply-side platform sends your data to an exchange—like a stock market for ads. Dozens of advertisers bid for the chance to show you their message. Highest bidder wins. Total time: under 200 milliseconds.

Now imagine the same system, but for content. Instead of bidding to show you ads, publishers vie to inform AI responses.

The AI uses quality metrics, not money, to determine winners. Publishers compete on relevance, accuracy, freshness, and authority. The signals that make content useful. This is PageRank for real-time AI responses—algorithmic evaluation operating in milliseconds rather than batch processing.

Here’s how it would work in practice : you ask Gemini: ‘What are the reviews of the new Google Pixel phone?’ That query broadcasts to participating publishers: tech reviewers, consumer sites, electronics retailers. They submit their best content to the auction. Gemini evaluates quality, recency & relevance, then synthesizes the winning submissions into your answer.

Notice what disappears : the demand-side platform. No more advertisers optimizing for clicks. Just publishers competing to be the most useful source. The internet gets fewer ads, but every piece of content fights for attention in an auction measured in milliseconds.

Publishers get traffic & attribution when selected, creating indirect revenue through brand building & subscriptions—but only if their content consistently wins on merit.

2025-07-22 08:00:00

OpenAI receives on average 1 query per American per day.

Google receives about 4 queries per American per day.

Since then 50% of Google search queries have AI Overviews, this means at least 60% of US searches are now AI.

It’s taken a bit longer than I expected for this to happen. In 2024, I predicted that 50% of consumer search would be AI-enabled.

But AI has arrived in search.

If Google search patterns are any indication, there’s a power law in search behavior. SparkToro’s analysis of Google search behavior shows the top third of Americans who search execute upwards of 80% of all searches - which means AI use isn’t likely evenly distributed - like the future.¹

Websites & businesses are starting to feel the impacts of this. The Economist’s piece “AI is killing the web. Can anything save it?” captures the zeitgeist in a headline.

A supermajority of Americans now search with AI. The second-order effects from changing search patterns are coming in the second-half of this year & more will be asking, “What Happened to My Traffic?”

AI is a new distribution channel & those who seize it will gain market share.

2025-07-20 08:00:00

In working with AI, I’m stopping before typing anything into the box to ask myself a question : what do I expect from the AI?

2x2 to the rescue! Which box am I in?

On one axis, how much context I provide : not very much to quite a bit. On the other, whether I should watch the AI or let it run.

If I provide very little information & let the system run : ‘research Forward Deployed Engineer trends,’ I get throwaway results: broad overviews without relevant detail.

Running the same project with a series of short questions produces an iterative conversation that succeeds - an Exploration.

“Which companies have implemented Forward Deployed Engineers (FDEs)? What are the typical backgrounds of FDEs? Which types of contract structures & businesses lend themselves to this work?”

When I have a very low tolerance for mistakes, I provide extensive context & work iteratively with the AI. For blog posts or financial analysis, I share everything (current drafts, previous writings, detailed requirements) then proceed sentence by sentence.

Letting an agent run freely requires defining everything upfront. I rarely succeed here because the upfront work demands tremendous clarity - exact goals, comprehensive information, & detailed task lists with validation criteria - an outline.

These prompts end up looking like the product requirements documents I wrote as a product manager.

The answer to ‘what do I expect?’ will get easier as AI systems access more of my information & improve at selecting relevant data. As I get better at articulating what I actually want, the collaboration improves.

I aim to move many more of my questions out of the top left bucket - how I was trained with Google search - into the other three quadrants.

I also expect this habit will help me work with people better.

2025-07-18 08:00:00

What happens when technology evolves faster than your sales process can adapt?

The last fifteen years, startups focused on building software around very well understood processes. We had built an assembly line for software sales, SDR to AE to customer success manager. We calculated ratios between these three total cost of sales and drove the factory to ever improved yields.

AI is upending all of that.

The underlying workflows are changing so quickly, software buyers no longer know what the ideal processes are, much less which is the best software to buy.

Model capabilities have evolved at 10x improvements every two years. Users are grappling to understand how to take advantage of these advances while boards are pressing teams to adopt AI.

A combination of all these factors has led to a reinvention of customer success : the forward deployed engineer.

Forward deployed engineers (FDEs) are the new customer success managers, the new solutions architects. They spend their time working with customers, understanding business challenge, and using technology to solve them - selling usage & outcomes.

In a software sales environment where buyers seek education, the underlying technology is advancing very quickly and there’s no stability. There’s no surprise that this role has become critical.

OpenAI has offered consulting services as well as Anthropic for custom enterprise deployments. Anthropic builds specialized enterprise implementation teams. Sierra employs agent engineers.

Palantir created this model. Their core insight, success comes from delivering outcomes on some software platform is now the standard for mid-market and enterprise software. The costs simply don’t justify themselves below price points of $100,000 or less per contract. Staffing a FDE costing $200k for a $10k contract - the math doesn’t work.

These forward-deployed engineers take the core platforms of AI and then mold them and tune them to work, defining new ways of building sales and marketing. Marketing and engineering teams - for example, agent managers.

The ability for customer success managers of the future to vibe code new platforms to deliver success on a basic platform is real . And it will be a requisite for these teams in an age where customer expectations of delivering value are shorter than ever.