Where the $435B AI market is actually making money? | a16z on AI pricing wars: what startups get wrong.
$175M seed valuations the new normal? & What 12,000 VC careers reveal about making partner.
đ Hey, Sahil here - Welcome back to Venture Curator, where we explore how top investors think, how real founders build, and the strategies shaping tomorrowâs companies.
Big idea + report of the week :
Are $175M seed valuations the new normal?
Do analysts actually become partners?: What 12,000 VC careers reveal about making partner.
Frameworks & insightful posts :
Where is the $435b AI market really making money?
a16z on AI pricing wars: what AI startups get wrong about enterprise pricing?
The unit economics Excel sheet template every founder should use.
FROM OUR PARTNER - SEMRUSH
đ If you had your competitorsâ growth playbook, what would you do?
With Semrush, you basically do.
See where competitors get their traffic from
Identify their top-performing pages and keywords
Track their presence across search and AI
Find gaps you can capture immediately
Itâs like having a live dashboard of your market. Use it to move faster, smarter, and with more confidence.
Good marketing is creative. Great marketing is informed.
Get full access for free for 7 days â
đ€ PARTNERSHIP WITH US
Get your product in front of over 115,000+ audience - Our newsletter is read by thousands of tech professionals, founders, investors and managers worldwide. Get in touch today.
START WITH
đ§ Big idea + report of the week
Are $175M seed valuations the new normal?
Itâs starting to feel familiar again. The speed. The pricing. The conviction at the very top of the market.
And the data now backs that feeling.
Peter Walker, Head of Insights at Carta, recently shared updated numbers showing that top 5% seed valuations have surged to ~$175M - nearly 3x growth in just 12 months. Even for a market thatâs used to cycles, this kind of acceleration stands out.
But whatâs more interesting isnât just how high valuations have gone.
Itâs how unevenly theyâre distributed.
At the very top of the market, things look almost surreal:
$100Mâ$175M seed rounds are getting done in days, sometimes hours
Top-tier founders (especially in AI) are raising pre-product or early traction
Large funds are aggressively moving into seed, compressing timelines and pushing prices up
Valuations are being set based on perceived future dominance, not present metrics
This is the part of the market everyone sees and reacts to. But step back, and a very different picture emerges.
The broader seed market hasnât moved nearly as dramatically.
Median valuations are still relatively stable and disciplined
Most companies are still raising based on traction, not narrative
Capital is more selective outside the top tier
What this creates is a massive bifurcation.
The âaverageâ seed valuation has effectively become meaningless because it blends two completely different markets:
A small group of AI-driven outliers is pulling valuations sharply upward
A much larger base of companies operating under normal constraints
And the gap between these two is widening faster than ever.
Walker himself describes it as having âa whiff of 2021-era ridiculousness.â
The pattern does feel similar:
Mega funds are pushing earlier into the stack
Deals are closing faster than diligence cycles
Secondary activity increasing
Narrative and momentum are playing a larger role in pricing
But thereâs one key difference this time. The underlying technology, AI, is actually delivering real traction.
And thatâs what makes this cycle more complex. Because both of these things can be true at the same time: AI is a genuinely transformative shift, and the capital flowing into it can still be mispriced.
That duality is what founders and investors need to understand. The biggest risk right now isnât just overvaluation. Itâs misreading the market youâre in.
If youâre a founder, the takeaway isnât âraise at $100M+, or youâre behind.â
Itâs understanding that:
Those deals exist - but theyâre outliers
They come with extremely high expectations
And they often require growing into a narrative that hasnât played out yet
If youâre an investor, the question shifts as well. Itâs no longer just about getting access. Itâs about asking:
Are you underwriting real outcomes or just momentum?
Is this valuation justified by category potential or driven by competition?
And most importantly, are you investing in a market⊠or in a moment?
Because cycles like this donât break evenly. They correct at the edges first. And right now, the edge of the market is exactly where the most capital is flowing.
Do analysts actually become partners, or is the VC career path broken from the start?
Most people trying to break into venture think the safest path is obvious: start as an analyst, work your way up, and eventually make partner.
But the data suggests something very different.
In one of the largest studies of VC careers ever conducted, Stanfordâs Ilya Strebulaev analysed 12,600+ venture professionals over nearly three decades. And the conclusion is uncomfortable for anyone planning a traditional path into VC:
The biggest predictor of making partner isnât skill, pedigree, or even deal success.
Itâs where you enter the industry. If you start too junior, the odds are stacked against you from day one.
The data shows that professionals who enter VC at the analyst or associate level are 70-80 percentage points less likely to become partners compared to those who join at mid-level roles like VP or Principal.
Thatâs not a small gap. Thatâs structural.
Which means the âstart early and climbâ narrative is often misleading. In reality, many partners donât grow up inside VCâthey enter it laterally after building credibility elsewhere.
And when you look at what actually increases your odds, a pattern starts to emerge. Operational experience matters more than almost anything else.
People who have worked at VC-backed startups - whether as founders, CXOs, or even early employees are significantly more likely to make partner. Theyâve seen company building firsthand, understand decision-making under uncertainty, and bring context that pure investors often lack.
In contrast, purely financial or academic paths donât perform as strongly as people expect.
MBAs increase your chances of getting promoted
But they actually correlate with worse investment outcomes on average
Non-MBA advanced degrees (MS/PhD) perform better across both promotion and deal success
Only top-tier MBAs (Stanford, HBS) seem to offset this gap
This creates a strange dynamic inside VC firms. The people most likely to get promoted arenât always the ones making the best investment decisions.
And the ones making strong investment decisions often didnât follow the âcleanâ path into venture. Thereâs another layer here thatâs harder to ignore.
Gender remains the only consistently negative predictor in the dataset. Female investors are significantly less likely to make partner - even after controlling for background, experience, and performance. Itâs not a pipeline issue. Itâs structural.
Put all of this together, and the VC career path looks very different from how itâs usually portrayed. Itâs less of a ladder and more of a filter.
Starting too early can actually limit long-term upside
Real-world operating experience compounds your advantage
Credentials help with access, but not necessarily with outcomes
And structural biases still shape who ultimately makes it
For anyone thinking about entering a venture, the takeaway isnât âdonât start.â Itâs understanding when and how to enter.
Because the fastest way to become a partner might not be getting into VC early.
It might be building something valuable outside it first and then coming in when you actually have leverage.
SOMETHING MORE
𧩠Frameworks & insightful posts
Where is the $435b AI market really making money?
For the past two years, the narrative around AI has been simple: this is the biggest tech wave since the internet.
More companies, more products, more users, more funding.
But if you zoom out and ask a different question - where is the money actually going?, The answer is surprisingly concentrated.
Apoorv Agrawal from Altimeter recently shared his analysis of the AI stack, and itâs counterintuitive. Despite all the innovation happening at the application layer, most of the value is still captured elsewhere.
Two years ago, the AI value chain looked inverted. The compute layer (semiconductors) captured the overwhelming majority of both revenue and profit, while applications - despite being closest to users - captured very little.
The expectation was that this would eventually flip, just like it did in previous tech cycles, such as cloud.
But two years later, even after massive growth, the structure hasnât changed much.
The AI ecosystem grew ~5x (from ~$90B â ~$435B)
Yet semiconductors still capture ~70% of total revenue
And ~79% of total gross profits
So while everything is growing, the distribution of value remains heavily skewed.
To understand this better, break the stack into three layers.
At the bottom is semiconductors (~$300B), which is essentially dominated by NVIDIA. Their data center business alone has scaled to a massive run rate, and when you add players like Broadcom and memory providers, the concentration becomes even clearer. NVIDIA alone controls roughly ~80% of this layer.
In the middle is infrastructure (~$75B), cloud providers like AWS, Azure, and Google Cloud, along with newer inference players. This is the most competitive layer, with value distributed more evenly across players.
At the top is applications (~$60B), where most founders are building. But even here, the market is highly concentrated. OpenAI and Anthropic alone account for a large portion of total revenue, with the rest spread across coding tools and emerging AI-native startups.
Whatâs more striking is not just revenue, but profitability.
Semiconductors run at ~73% gross margins
Infrastructure at ~55%
Applications at ~33%
When you translate that into actual dollars:
Semis generate ~$225B in gross profit
Infra generates ~$40B
Apps generate only ~$20B
In other words, the layer with the most innovation is capturing the least profit.
This is why Apoorv summarises it simply: The most profitable strategy in AI today is still selling the shovels.
Another important insight is how growth is being distributed.
Yes, the application layer is growing the fastest in percentage terms (roughly 10â12x in two years). But in absolute terms, the majority of value creation is still happening in semiconductors.
Semis added ~$225B in new revenue
Apps added only ~$55B
NVIDIA alone added ~$175B
Thatâs nearly 3x the size of the entire application layer today.
This also explains the massive capex cycle weâre seeing. Hyperscalers are spending aggressively to secure compute:
~$443B in capex in 2025
Expected to exceed ~$600B in 2026
~75% of that (~$450B) is going into AI infrastructure
Theyâre effectively racing to control supply in a market where demand is exploding faster than capacity.
At the same time, every major player is trying to reduce dependence on NVIDIA by building custom chips - Google TPUs, Amazon Trainium, Microsoft Maia, Metaâs internal chips. But so far, none have fully matched NVIDIAâs dominance at scale.
So what happens next?
The long-term expectation is still that value shifts upward - from infrastructure to applications - just like it did in the cloud era.
But the timeline may be much longer than people expect.
Cloud took ~15 years to flip from hardware â software dominance
AI may follow a similar trajectory
At the current pace, it could take over a decade for apps to capture the majority of profits
For that to happen, two things need to accelerate:
Compute needs to get cheaper (compressing margins at the bottom)
Applications need to capture more differentiated value (expanding margins at the top)
Both are happening - but not fast enough yet.
So, especially for founders - AI may feel like an application-driven revolution. But economically, itâs still a compute-driven market.
Which means:
The biggest profits today sit at the bottom of the stack
The most competition sits in the middle
And the biggest long-term opportunity still sits at the top
But getting there will take time. Because in every platform shift, the early winners sell infrastructure. The enduring winners build applications that capture value later.
And right now, weâre still very early in that transition.
a16z on AI pricing wars: what AI startups get wrong about enterprise pricing?
Everyone assumes the same thing right now: AI is getting crowded, products look similar, and the only way to win enterprise deals is to go cheaper.
That assumption feels logical. Itâs also quietly wrong.
Tugce Erten recently broke this down after speaking with multiple enterprise buyers, and the insight is uncomfortable but important - most AI startups are not losing because theyâre too expensive. Theyâre losing because they havenât become indispensable enough yet.
And once you see that, the entire pricing conversation changes.
One of the biggest misconceptions founders carry into enterprise sales is that budgets are tight and buyers are extremely price-sensitive. But whatâs actually happening inside companies looks very different.
Many large enterprises already have meaningful AI budgets allocated, and instead of choosing one tool, theyâre often deploying multiple tools for the same workflow. This isnât confusion - itâs strategy.
They are hedging against model volatility, testing performance across vendors, and avoiding dependence on a single provider in a fast-moving market.
Which means in many cases, youâre not competing to win the budget - youâre competing to survive the eventual consolidation.
Because when that happens, teams donât keep three tools. They keep one. And the one they keep usually isnât the cheapest. Itâs the one that:
works reliably under real conditions
integrates deeply into existing workflows
improves fast based on feedback
feels like itâs actively being built for them
This is why premium pricing still exists in AI, even in a crowded market.
If your product is genuinely better or even clearly perceived to be better, you can often charge a 10-20% premium without meaningful resistance. Some buyers even structure usage this way: premium tools for high-stakes workflows, cheaper ones for simpler tasks.
But this advantage doesnât last automatically.
In AI, perception shifts fast. A new product with better outputs, cleaner UX, or stronger distribution can reset expectations within months. So premium pricing isnât something you âhave.â Itâs something you keep earning.
What most founders underestimate is that pricing isnât just about the number - itâs about the structure.
How you charge often shapes the entire buying conversation:
per-seat pricing â easy to compare â pushes price competition
per-outcome pricing â harder to compare â shifts focus to value
Consumption pricing â reframes discussion around usage efficiency
Enterprise buyers are already pushing for this shift. They want pricing that reflects value, but they also need predictability for planning.
So the smartest companies arenât choosing one model - theyâre offering flexibility. Fixed pricing for teams that need certainty, and performance-linked models for teams that want tighter value alignment.
That flexibility often wins deals without touching the headline price.
Another place where most teams go wrong is during the sales process itself.
Deals donât usually fail because of the final contract value. They fail because getting started feels heavy - long procurement cycles, security reviews, internal approvals.
So instead of discounting the product, the smarter move is to reduce the cost of entry.
You see this playbook everywhere now:
generous credits or usage during POCs
expanded free tiers that act like acquisition channels
over-delivering early to prove real value fast
The goal isnât to win on price. Itâs to win on adoption - to get embedded before the market consolidates.
But the most important shift is this:
The real price war isnât with your competitors. Itâs with your customerâs own engineering team.
As model costs fall and APIs get easier, companies start asking a different question - not âwhich vendor is better?â but âcan we just build this ourselves?â
And increasingly, the answer is yes. Thatâs where shallow products break.
Because if your product is easy to replicate internally, pricing pressure never really goes away. It just gets delayed.
The only real defence is building something expensive to rebuild. In practice, that usually means:
deep workflow integration into day-to-day operations
accumulated data and context that improve outputs over time
domain-specific tuning that generic models canât match
continuous iteration based on real customer usage
If you step back, the pattern becomes clear.
Enterprise buyers arenât rewarding the cheapest product. Theyâre rewarding the one that proves itself in real workflows, reduces risk, and becomes difficult to remove.
Which means the real question for founders isnât âhow do we win this deal on price?â Itâs whether your product is strong enough that price stops being the deciding factor.
Because once that happens, youâre no longer in a price war. Youâre in a position of leverage.
The unit economics Excel sheet template every founder should use.
Most founders talk about growth. Few can clearly show how each new customer makes or loses them money. Thatâs why running a unit economics model early matters; it tells you if your business will compound or collapse.
Hereâs how to use the sheet Iâve shared:
1. Map your revenues
Enter your average revenue per unit (subscription, fee, or transaction).
Decide if itâs recurring (monthly/annual) or one-time.
Pro tip: if you donât know exact numbers yet, start with estimates and refine over time.
2. Add cost of sales (COGS)
Enter costs as a % of revenue (hosting, delivery, support).
The sheet calculates your gross margin per unit automatically.
3. Plug in churn + growth assumptions
Set billing cycle (monthly = 1, annual = 12).
Add churn % (customers who drop each cycle).
Add expected annual margin growth (e.g., 5%).
The model uses these to estimate the average customer lifetime.
4. Add acquisition + retention costs
Enter your CAC (cost to acquire one customer).
Add optional retention/expansion costs if relevant.
5. Review key outputs
LTV (Lifetime Value): how much a customer is worth over their lifecycle.
LTV/CAC ratio: >3 is usually healthy.
CAC payback period: months it takes to earn back the acquisition spend.
Cumulative cash flows: show when you turn profitable per customer.
6. Stress-test scenarios
Increase churn by 30%.
Drop your AOV (average order value) by 20%.
Raise CAC by 50%.
See how fast LTV/CAC breaks. This is where most founders get surprised.
Why this matters
Investors use this as a quick sanity check.
It forces you to confront pricing, churn, and acquisition head-on.
A business with shaky unit economics will break no matter how good the growth story sounds.
Download the sheet, plug in your numbers, and run 3 scenarios: best case, expected case, worst case. Youâll instantly see whether your idea scales or needs fixing.
NEWS RECAP
đïž This week in startups & VC
New In VC
Accel, a global venture firm, has raised $5B in fresh capital to back late-stage startups, with $4B allocated to its Leaders Fund and $650M to a sidecar vehicle. (Link)
AâStreet, a Bentonville, Ark.-based multi-stage investment fund focused on improving PK-12 student learning and achievement, announced a $675m capital raise. (Link)
Eclipse, a Palo Alto, CA-based venture capital firm, raised $1.311 billion for two funds. (Link)
Futurepresent, a NYC, Berlin, and Munich, Germany-based venture capital firm backing AI startups in the US and Europe, launched with its first $300M vehicle. (Link)
Zero Shot, a U.S.-based VC fund founded by former OpenAI leaders, has raised $20M in its first close toward a $100M target to invest in early-stage AI startups. (Link)
New Startup Deals
Traza, a NYC-based AI-native procurement platform, raised $2.1M in Pre-Seed funding. (Link)
SolvaPay, a Stockholm-based AI payments platform, raised âŹ2.4M in Pre-Seed funding. (Link)
Helical, a London, UK-based virtual AI lab for pharma, raised $10M in Seed funding. (Link)
ActionAI, a NYC-based enterprise AI automation company, raised $10M in Seed funding. (Link)
DeepCyte, a Wilmington, DE & Copenhagen-based techbio company, raised $1.5M in Seed funding. (Link)
Flashpass, a Columbus, OH-based digital skills platform, raised $4.25M in Seed funding. (Link)
TODAYâS JOB OPPORTUNITIES
đŒ Venture capital & startup jobs
All-In-One VC Interview Preparation Guide: With a leading investor group, we have created an all-in-one VC interview preparation guide for aspiring VCs. Donât miss this. (Access Here)
Analyst, Global Investment Team - 500 Global | USA - Apply Here
Associate, Data Operations - Iconiq Capital | USA - Apply Here
Finance Associate - RA Capital | USA - Apply Here
Fund Controller - NFX | USA - Apply Here
Vice President, Investor Relations - General Atlantic - Apply Here
PE & VC Partner Manager - Dealhub | UK - Apply Here
Partner 22 -a16z | USA - Apply Here
Infra / Platform Engineer - Pear VC | USA - Apply Here
Associate / Senior Associate - Stepstone Group | Italy - Apply Here
Investment Analyst - Lunicorn Venture | UK - Apply Here
Investor (Senior Associate/Principal) - Square Peg | USA - Apply Here
Program Manager - a16z | USA - Apply Here
Chief of Staff - Greycroft | USA - Apply Here
Associate - Aditum Bio | USA - Apply Here
Associate or Senior Associate - AI - BVP | USA - Apply Here
Venture Fellow - age1 | USA - Apply Here
Investment Associate - 500 Global | USA - Apply Here
PARTNERSHIP WITH US
Get your product in front of over 115,000+ audience - Our newsletter is read by thousands of tech professionals, founders, investors and managers worldwide. Get in touch today.
đŽ Share Venture Curator
You currently have 0 referrals, only 5 away from receiving a đ gift that includes 20 different investorsâ contact database lists - Venture Curator