Actionable guides, real cases, and strategic frameworks to help you build, grow, and fund your Web3 project. From tokenomics and roadmaps to launch and beyond.
Ten months. One million users. $43 million raised. $185 million valuation. Phia, the AI shopping agent co-founded by Phoebe Gates and Sophia Kianni, launched in April 2025 and closed a $35 million Series A in January 2026, led by Notable Capital with participation from Khosla Ventures and returning investor Kleiner Perkins. For a company with 20 people and no traditional marketing budget, these numbers don't happen by accident.
This is a breakdown of how they did it, the product decisions, GTM choices, distribution mechanics, and business model that made it work.
Gates and Kianni started with personal frustration.
Both avid secondhand shoppers, they realized that buying clothes online meant opening dozens of tabs, cross-referencing prices, guessing at resale value, and ultimately hoping they were making a reasonable decision. "The way people shop online hasn't really changed since Amazon launched," Kianni told Glossy. "You can get anything instantly, but the actual experience hasn't evolved."
The global e-commerce apparel market is projected to reach $1,160.56 billion by 2030, growing at a CAGR of 8.6% from 2022 to 2030. The US secondhand apparel market grew 14% in 2024. Despite that volume, the shopping interface: search, filter, buy was unchanged from 2005.
Their positioning was clear before the product was built: "Google Flights for fashion." One sentence that communicated the product to any target customer who had ever booked a flight and wondered why shopping wasn't that simple.
Phia is a mobile app and browser extension. It does five things in real time:
The product database covers 350 million items. The company processes millions of searches daily and ingests hundreds of millions of new products each day. A proprietary search model rolled out post-launch reduced latency by 80% and increased monetized GMV by 40%.
The pivot that mattered: The first version was a desktop Chrome extension focused on secondhand comparisons. User feedback showed their target customer shopped on mobile, and cared more about instant price comparison than secondhand discovery. They rebuilt for mobile and reoriented the core use case. That adjustment happened before the seed round.
Phia's GTM is one of the most studied examples of founder-led, zero-paid-media growth in recent consumer tech. It ran on four mechanics simultaneously.
Gates and Kianni built audience infrastructure before the product launched.
Why this works as GTM:
The key insight: Growing an audience translates to growing a user base. After all, it's the same group of people they are trying to reach. Most founders build an audience after reaching product-market fit. Phia built it before launch.
The seed round is worth examining not just for the capital but for what the investor list was designed to do.
Each investor was chosen for network effect, not just check size. Jenner and Blakely appeared on the podcast as investors. The round itself generated press that drove app downloads. The investor list was a distribution channel.
Soma Capital found them through LinkedIn before they started fundraising. Kleiner Perkins came through an introduction from Soma. The fundraising process reflected the same logic as their product: warm referrals and community over cold outreach.
Phia ran a bi-weekly session inviting 30 women to shape product features. This served two functions: genuine product research and building evangelists before scale.
Users who participate in shaping a product are significantly more likely to share it. Phia converted their early users from customers into contributors, which compressed the feedback loop and generated word-of-mouth that paid acquisition couldn't replicate.
The product feedback sessions also surfaced the mobile pivot. The desktop-first assumption was wrong. They found out from 500 early users before spending a year building the wrong distribution format.
Kim Kardashian filmed a teaser for Phia's launch in early April 2025. It ran on Instagram before the app went live. Kianni and Gates were already known names: Kianni as one of the youngest UN advisers and founder of Climate Cardinals; Gates by association with her father, who himself joined the Phia customer service team for a day and posted about it. Bill Gates working a shift at his daughter's startup is the kind of coverage that no marketing budget produces.
Phia runs on performance-based affiliate revenue. When a brand makes a sale through Phia, the app takes a cut. There is no upfront fee for brand partners.
This zero-cost-to-join model was deliberate. "A lot of our partners were very much taking a bet on us and joining the platform when we had less proof points," said Kianni. Lowering the barrier to entry let them onboard 5,000 brands in the first five months without a sales team. By January 2026, they had proof points to show:
These numbers changed the brand conversation from "trust us" to "here's what your category peers are seeing." The zero-upfront model stopped being a concession and became a competitive position. Revenue model evolution:
The business is structured as a two-sided marketplace: consumers get better purchasing decisions, brands get lower-cost customer acquisition. Both sides win at the transaction level. The model is self-reinforcing: more users improve brand partner data, which improves recommendations, which improves conversion, which attracts more brands.
The Series A press release introduced a phrase worth examining: "AI alignment layer between consumers and brands." This is a category creation attempt.
Hans Tung of Notable Capital framed the investment thesis directly: "Historically, shopping was built for an internet of pages and filters. Now, AI sits between people and products, and the challenge is no longer access. It is understanding intent, taste, and trust in real time."
Phia is positioning itself as the infrastructure layer, the intelligence that sits between what consumers want and what brands have, operating at the moment of decision rather than before or after it. The roadmap reflects this:
The word "agent" appears throughout. The product is moving from a comparison tool to an autonomous shopping assistant, one that knows your preferences, tracks your budget, monitors prices, and acts on your behalf. That's a different product category than price comparison, and Phia is making the vocabulary shift before the product is fully built.
Phia nailed its go-to-market by making a handful of sharp decisions early, before scale locked them in. Clear positioning, low-friction supply, and a fast product pivot aligned distribution with real user behavior.
Phia made four decisions, or failed to make four decisions, that created real and compounding risk. None of them are fatal yet. All of them were avoidable.
The most useful reading of Phia is as a real-time case study in what happens when a smart GTM meets the friction of scale. Either way, the questions they raise are ones every founder building a consumer product with broad permissions, a founder-dependent distribution strategy, and an affiliate revenue model should be asking now before they're facing them at Series A.
Phia is a story about sequencing. Gates and Kianni made the right decisions in the right order: audience before product, mobile pivot before scale, performance-based brand model before they had data to justify upfront fees. Every element of the GTM, the podcast, the investor list, the community sessions, the pre-launch PR served more than one function simultaneously. Nothing was decorative.
The $185 million valuation is the outcome. The mechanism was simpler: find the target customer, become someone they already trust, and remove every barrier between them and the product. Then do the same thing for the brands on the other side.
Most startups get one of those right. Phia got both, before the Series A, with 20 people and no paid media budget. That is the case study.
Q: Can the founder-led media strategy work if your founders aren't already public figures?
Gates and Kianni had name recognition, but the mechanism was consistency and specificity. The Burnouts worked because it targeted the exact audience Phia needed, documented a real journey, and launched before the product. A niche founder with 20,000 highly aligned followers will outperform a generalist with 500,000. The question is whether they're willing to build in public and whether their natural audience matches their target customer.
Q: Why did the zero-upfront brand model work, and when does it stop working?
It works at the start because it removes the only objection that matters before you have data: "Why should we trust you?" Once you have conversion metrics, return rate data, and GMV figures, the conversation changes entirely. The risk is that a pure affiliate model underprices your value early. Phia solved this by treating the zero-fee period as a data acquisition strategy. They moved to dashboard tools and SaaS-adjacent revenue once the proof existed.
Q: What's the real lesson from the Phia mobile pivot, and how do you apply it before you have 500 users?
The lesson is that they ran structured feedback sessions with 30 users every two weeks before launch, which meant bad assumptions surfaced before they became expensive. Most founders skip this because it feels slow. Phia found out their entire distribution format was wrong before the seed round. The same discovery post-Series A would have cost them 12 months and most of their credibility.
Q: Is the Phia GTM model only viable with venture backing?
Phia raised a pre-seed before launching, but the core mechanics: founder content, community sessions, performance-based supply onboarding, cost almost nothing. The bi-weekly user sessions required a calendar invite. The affiliate model required no sales team. The venture capital accelerated scale, but none of the early growth levers required it. The harder dependency is founder time and willingness to be visible, neither of which money solves.
.png)
Phia went from a Stanford dorm room to a $185M valuation in under a year. A deep GTM analysis of founder-led marketing, zero-cost distribution, performance-based brand partnerships, and AI product strategy.Read more
75% of venture-backed startups fail, and 42% of them fail because they built something nobody wanted. The team did not think hard enough about who would buy it, how to reach them, and why they would pay.
A go-to-market strategy is the plan that answers those three questions before you spend your budget finding out the hard way.
This guide covers what a GTM strategy actually includes, how to build one that works in a competitive market, and what the companies that got it right actually did.
A go-to-market strategy is a comprehensive plan for bringing a product or service to market. It defines your target audience, your value proposition, how you'll reach potential customers, what you'll charge and how your sales and marketing teams will work together to drive adoption.
A well-defined GTM strategy requires you to make concrete decisions before you launch: who the ideal customer is, what problem you're solving for them, which distribution channels you'll use, and what pricing strategy makes sense for the market you're entering. Without a GTM strategy, you're guessing. With one, you're testing a hypothesis.
The pattern repeats across industries. Founders build the product, then figure out the market. By the time they realize their ideal customer doesn't actually have the problem they assumed, the budget is gone.
Research from CB Insights puts "no market need" as the primary reason for startup failure, cited by 42% of failed founders. A strong GTM strategy forces you to validate market demand before you commit resources to a full launch.
The alternative, bringing a product to market without a plan, tends to produce one of two outcomes: either the product never finds its target customers, or it finds the wrong ones and can't scale.
A go-to-market plan also creates alignment across the team. When sales, marketing, and product agree on who the target customer is, what the value proposition says, and which channels the go-to-market team will prioritize, execution gets faster and cheaper.
The first step in any go-to-market strategy is defining exactly who you're building for. The target customer question has two levels.
In B2B go-to-market contexts, these are often different people. The economic buyer controls the budget but may never use the product. The end user experiences the pain every day but may not control the budget. Your GTM strategy needs to account for both.
Slack's early GTM strategy targeted teams inside companies rather than companies themselves. Individual users adopted the product, then brought it into their organizations. The target customer was the team lead tired of email. That choice drove everything else about how they distributed and priced the product.
Your value proposition is the specific outcome your target customer gets from your product or service. It should be concrete enough that your target audience recognizes themselves in it.
Vague value propositions ("we help teams work better") don't convert. Specific ones do ("your team goes from 30-minute standup meetings to 10-minute async updates").
Positioning connects your value proposition to the competitive market. Where do you sit relative to alternatives? What makes your product different in a way that matters to your target customer? Why would someone choose your product over the existing solution they're already using?
The best positioning, as discussed in this article, is the kind that competitors can't copy without hurting themselves. Mistral positioned itself as the open-source European sovereignty alternative to OpenAI, a position OpenAI structurally couldn't match. That's positioning with teeth.
Pricing is a GTM decision. The price you charge signals who the product is for, whether it's positioned as a tool or an investment, and what kind of customer relationship you're creating.
The right pricing strategy depends on your total addressable market, the size of your average deal, how long your sales cycle is, and whether you're building a self-serve or sales-led motion.
Distribution is often where GTM strategies break down. A product can have a clear target audience and a compelling value proposition and still fail if it never reaches the people who would buy it.
Distribution channels are the paths through which your product or service reaches your target customer. They include direct sales, partnerships, product-led growth (the product itself drives adoption), content and SEO, paid acquisition, and community.
The right channels for your GTM strategy depend on who your target customer is and how they make purchase decisions. Enterprise buyers don't find software through Instagram. Developer tools spread through GitHub and Hacker News. Consumer apps grow through app stores and word of mouth.
Airbnb's early distribution came from Craigslist. They built a mechanism that let hosts automatically repost their Airbnb listings on Craigslist, reaching an existing audience already looking for accommodations. Professional photos increased booking rates by 40%. Geographic market-by-market expansion built critical mass in individual cities before going broader. The result: Airbnb's referral program produced 900% year-over-year growth in first-time bookings across emerging markets.
How your sales team is organized depends on your GTM motion. Product-led growth companies often have smaller sales teams that focus on converting high-usage free accounts and closing larger enterprise deals. Sales-led companies invest earlier in sales headcount and outbound motion.
For B2B go-to-market strategies, the sales process needs to match the complexity of the purchase. Simple products with clear ROI can close quickly with minimal sales involvement. Complex enterprise products with multiple stakeholders require a longer cycle with different messages for different roles.
One thing that shows up consistently in B2B go-to-market analysis: a sales strategy that doesn't account for the full buying committee tends to stall. The person who finds the product isn't always the person who signs the check.
A go-to-market plan without a marketing strategy is a plan for waiting. Marketing creates awareness, builds trust with potential customers before the sales conversation starts, and handles the education work that would otherwise fall on the sales team.
Customer acquisition cost (CAC) and lifetime value (LTV) are the metrics your marketing strategy ultimately has to optimize. Knowing what it costs to acquire a customer relative to what they're worth tells you which channels to invest in and which to abandon.
The marketing strategy needs to align with the distribution channels and the target customer. Content marketing works when your target customer searches for solutions to their problem. Community works when your product solves problems people discuss publicly. Paid acquisition works when the unit economics support it and the audience is targetable.
A comprehensive GTM plan follows seven steps. Each one builds on the last, skipping steps tends to produce a strategy that looks complete on paper but breaks down in execution.
A great go-to-market strategy is specific. Trying to reach every potential customer at once means reaching none of them effectively. Start with the segment where you have the clearest advantage and the highest probability of success.
How long does it take to develop a go-to-market strategy?
It depends on the complexity of the product, the maturity of the market, and how much customer research has already been done. For most early-stage startups entering an existing market, developing a workable GTM strategy takes two to six weeks of focused work: customer interviews, competitive research, pricing decisions, and channel selection. Creating a go-to-market strategy for a genuinely new market category, where you're educating customers rather than converting them from existing solutions, takes longer, because the market research has to establish whether demand exists before the strategy can be built around it.
What makes a go-to-market strategy successful?
A successful GTM strategy starts with a validated understanding of customer needs. It targets a specific enough segment that messaging and channel choices can be optimized rather than spread thin. It aligns pricing with the value delivered to the target customer. It selects distribution channels based on how the target customer actually discovers and evaluates new products. And it defines success metrics clearly enough that the team can tell whether the strategy is working before the budget runs out. Most failed GTM strategies skip one of these steps.
What is a B2B go-to-market strategy and how is it different from B2C?
A B2B go-to-market strategy targets businesses as customers rather than individual consumers. The key differences are in the sales process, the decision-making structure, and the timeline. B2B purchases typically involve multiple stakeholders: an economic buyer, a technical evaluator, end users and longer sales cycles. B2B GTM strategies tend to emphasize relationship-building, ROI demonstration, and direct sales or partner channels. B2C GTM strategies focus more on mass reach, emotional resonance, and reducing friction in the purchase process. The targeting approach differs too: B2B GTM strategy relies heavily on firmographics (company size, industry, revenue) alongside individual buyer personas, while B2C GTM strategy focuses primarily on consumer demographics and behavior.
How do you measure whether a go-to-market strategy is working?
A GTM strategy should be judged by a few core metrics: customer acquisition cost, conversion at each funnel stage, time to first value, 30/60/90-day retention, and revenue from the target segment. If CAC rises while conversions fall, fix the channel or messaging. If retention is low, the product isn’t delivering value or you’re attracting the wrong customers. A plan built around these metrics can be quickly diagnosed and improved. One that isn’t, will burn the budget before anyone understands why.
.png)
Learn how to build a go-to-market strategy that actually works. Real cases from Slack, Airbnb, and Zoom. Covers target audience, pricing, distribution, and GTM execution for startups.Read more
In October 2018, USDT, the largest stablecoin by volume, traded at approximately $0.87 on major exchanges. USDC, in its early period, regularly traded above $1. Neither outcome was a crisis. Both were normal for what the market was at the time: few participants, thin liquidity, slow arbitrage, fragmented trading across exchanges with no efficient way to close price gaps.
By 2019–2023, according to BIS data, stablecoins traded within a tight band of $1 approximately 94% of the time. Deviations became smaller. Recovery became faster.
The tokens themselves had not changed in any fundamental way. What changed was the infrastructure built around them, who could access redemption, how many participants were willing to arbitrage, and whether the market trusted that a $1 redemption was real.
That is the argument this piece makes: a stablecoin peg is a market outcome. And market outcomes are built.
Peg stability is the result of three layers working simultaneously. Remove any one and the other two cannot compensate.
None of these layers works in isolation. A well-designed primary market with no arbitrageurs produces deviations that persist. A deep secondary market built on fragile trust collapses the moment that trust does. The peg holds when all three are functioning at once.
The primary market is where the economic logic of a stablecoin begins. Authorized participants can mint or redeem tokens directly with the issuer at $1. If the market price falls below that, someone buys at a discount and redeems for a dollar. The spread is the profit. In theory, this alone should keep the peg tight.
What the theory skips is access. Primary market access is gated. Minimum transaction sizes, onboarding requirements, and settlement timelines that can run from hours to days mean that the theoretical arbitrage mechanism is only available to a limited group.
Federal Reserve Board research confirms the consequence: even with fully backed reserves, restricted or slow redemption directly reduces arbitrage efficiency. Deviations persist longer because fewer participants can act on them.
The primary market creates conditions under which the market can stabilize the peg. The distinction matters for issuers: a redemption mechanism that works on paper but is inaccessible at speed or scale is a design feature that the market cannot use.
If the primary market sets the anchor, the secondary market is where the rope either holds or snaps. Price formation happens on exchanges and DEXs. When a stablecoin trades below $1, arbitrageurs buy it there and redeem with the issuer. The gap closes and the peg recovers.
How quickly that happens and how deep the deviation goes before it does depends on who is willing to act.
The USDT and USDC comparison from the 2021–2022 period makes this concrete. USDT averaged a secondary-market discount of around 55 basis points. USDC averaged approximately 1 basis point.
The difference in reserve quality or redemption mechanics between the two was not large enough to explain a 54-basis-point gap. The explanation sits elsewhere: USDT had roughly 6 active monthly arbitrageurs. USDC had more than 500.
More participants means smaller deviations, faster recovery, and less exposure to any single actor's willingness to trade on a given day. That is the structural finding and it points directly to what issuers need to build.
There is a second-order consequence worth naming. Research from the University of Chicago shows that more efficient arbitrage reduces deviations in normal market conditions but increases sensitivity to mass redemption events. Tighter arbitrage makes the system faster in both directions. Issuers should plan for that.
Breadth of listings compounds the participant depth dynamic. A stablecoin present across more venues and integrated into more use cases has more distributed liquidity. Localized imbalances, a spike in selling pressure on one exchange, get absorbed rather than amplified.
Reserve quality and transparency do not hold the peg mechanically. They determine whether market participants believe the peg is worth holding.
Tether's elimination of commercial paper from its reserves in 2022 illustrates this. The operational structure of the token did not change. What changed was the risk calculation for arbitrageurs and institutional participants who had been uncertain whether redemption would work as advertised under pressure. That shift in confidence had measurable market effects.
The BIS frames the underlying dynamic precisely: transparency's stabilizing effect is conditional on the starting level of trust. When trust is high, more disclosure stabilizes. When trust is fragile, the same disclosure can accelerate the reaction it was meant to prevent.
For issuers, this means the trust layer is a structural one. Participants will arbitrage only when they hold three beliefs at once: the redemption mechanism works, the token is genuinely worth $1, and the reserves can be liquidated fast enough to make that true under stress. All three must hold. A gap in any one is enough to produce hesitation and hesitation at the wrong moment is a depegging event.
The USDT and USDC history produces a clear conclusion: peg stability is not designed in at launch. It is built over time through deliberate infrastructure decisions. Four of those decisions matter most.
The history of stablecoin pegs is a story about market infrastructure catching up to a promise.
USDT made that promise in 2014. The market took years to develop the depth, the participants, and the operational maturity to keep it reliably. USDC built tighter conditions around the same promise faster, which is why its peg deviation was 54 basis points smaller despite operating in the same market.
The lesson is direct: the peg you announce at launch is not the peg you will have. What you will have is determined by how many participants can arbitrage, how quickly they can access the primary market, and whether they trust that the dollar on the other side of the trade is real. None of those conditions are set at issuance. All of them are built.
Stablecoin pegs break when arbitrage fails. That can happen because participants lack access to the primary market (mint/redeem), because liquidity on secondary markets is too thin to absorb selling pressure, or because trust in the issuer collapses and participants choose not to arbitrage even when the economics support it. All three layers need to function simultaneously. A failure in any one is enough to produce a depegging event.
The primary market is the direct channel between the issuer and authorized participants who can mint or redeem tokens at exactly $1. The secondary market is everywhere else: exchanges, DEXs, OTC desks, where price floats freely and must be pulled back toward $1 by arbitrageurs acting on the primary market anchor. The primary market sets the economic incentive. The secondary market determines whether enough participants can act on it fast enough to matter.
Research from 2021–2022 shows USDC had more than 500 active arbitrageurs per month. USDT had roughly 6. That difference in participant depth explains most of the gap between USDC's ~1 basis point average secondary-market discount and USDT's ~55 basis points. More participants means faster correction and less dependence on any single actor's willingness to trade on a given day.
No. Reserves are a precondition. Even fully collateralized stablecoins depeg when redemption access is slow or restricted, when too few participants can act as arbitrageurs, or when market confidence in the issuer breaks down. Reserve quality affects trust, trust affects participation, and participation is what holds the peg. The chain has multiple links and needs all of them.
Four things: broad liquidity distribution across CEXs and DEXs with dynamic incentive programs that respond to deviation; institutional arbitrage partners with defined KPIs covering spread, depth, and activity minimums; low barriers to entry for smaller arbitrageurs to prevent concentration risk; and a real-time monitoring system with pre-defined stress responses. Participant count is the strongest predictor of peg tightness, everything else serves that goal.
.png)
Stablecoin pegs are built over time through market infrastructure, arbitrage depth, and trust. Explore how USDT and USDC achieved peg stability, and what it means for new issuers.Read more
Nobody out-resourced OpenAI, Google, or Amazon. These three AI startups found a different way in and each one used a move the giant in front of them couldn't copy without hurting itself.
April 2023. Three researchers from DeepMind and Meta registered a company in Paris. One month later, they closed a €105M seed round. The timing is precise: ChatGPT has just redefined the market, European companies are looking for alternatives, and no credible option exists yet.
Mistral's bet was simple on paper and hard to execute. While OpenAI, Anthropic, and Google were racing to build the most powerful closed models, Mistral went open source under Apache 2.0 licenses. Because it was the one move a US-based closed model company couldn't match without unraveling its entire business model.
The strategy worked for a specific reason. European enterprises: banks, governments, defense agencies have hard constraints that American cloud AI can't satisfy. When Mistral says "deploy this on your own servers, the model is yours," that is the real product.
In late 2025, Mistral partnered with SAP and the French and German governments to build a sovereign AI stack for public administrations, ensuring government data is processed using EU-compliant technology. HSBC also chose Mistral as an AI partner for private cloud deployment, giving the bank flexibility, data security, and lower latency compared to cloud alternatives.
ASML, the Dutch semiconductor equipment giant, led a €1.3 billion Series C and took an 11% stake. Total funding since 2023 has crossed €2.8 billion. CEO Arthur Mensch said at Davos in January 2026 the company is on track to exceed $1 billion in revenue by end of year.
The takeaway: Mistral built the only model a specific set of buyers could actually use. The question worth asking before you start competing: is there a segment your main competitor is structurally excluded from? Not unwilling to serve, excluded, by their own business model or regulatory exposure?
Perplexity's founding pitch took about fifteen seconds to explain. Google gives you links, but we give you answers.
That insight wasn't obvious when the company started in 2022 with four people and a narrow hypothesis: search is broken for people who need to know things, and Google can't fix it without destroying its advertising business. Every link Google shows is a monetization opportunity. Switching to direct answers means removing the unit of ad inventory.
According to AI Funding Tracker, in January 2024, Perplexity was worth around $500 million. By December 2024, investors valued it at $9 billion. By September 2025, the number hit $20 billion – a 40x valuation jump in under two years, from a company with roughly 250 employees and no advertising budget.
Perplexity reached $200 million in annual recurring revenue by October 2025, representing 4.7x growth year-over-year. Growth came almost entirely from product quality and word of mouth.
The distribution strategy evolved intelligently. Samsung launched a Perplexity AI-powered TV app as part of Vision AI Companion, available on all 2025 Samsung TVs, with users receiving a free 12-month Pro subscription. Hardware partnerships replaced the sales team. Every device integration is an acquisition channel without acquisition cost.
The business model decision is worth examining separately. Perplexity tried advertising in 2024 and killed it. Executives concluded user trust is worth more than ad revenue, a direct contrast to how Google is structurally forced to operate. The company bet subscriptions and enterprise contracts over advertising, and the 4.7x revenue growth suggests the bet is working.
The takeaway: find the problem the market leader creates by existing. A structural limit they can't remove without damaging themselves. Then build the solution and wait for the constraint to become visible to everyone.
Two Polish engineers, Mati Staniszewski and Piotr Dąbkowski, started ElevenLabs in 2022 because they were frustrated watching poorly-dubbed American movies in Poland. The product they built first was simple: 30 seconds of audio, and a model clones the voice.
They gave it away for free and developers adopted it immediately. The free tier created a distribution network the company never had to build itself: creators embedded ElevenLabs into tools, apps, games, and workflows. By the time enterprise buyers showed up, the product was already everywhere.
The ARR trajectory tells the story directly, ElevenLabs hit $330 million in ARR in 2025, up 175% year-over-year. Enterprise clients include Deutsche Telekom, Revolut, Square, and the Ukrainian Government.
In February 2026, ElevenLabs raised $500 million in a Series D led by Sequoia Capital at an $11 billion valuation, more than tripling its valuation from one year prior.
Google and Amazon both have voice synthesis. OpenAI has audio capabilities. None of them chose to make voice AI the center of their research and product roadmap. ElevenLabs did and then distributed the product free to developers until the usage created its own enterprise demand.
The takeaway: the developer layer is the most efficient distribution channel in software. Bottom-up adoption is slower to start, harder to control, and faster to compound than direct sales. Freemium for developers is the cheapest sales force available.
Each of these AI startups found the one axis their main competitor couldn't match without hurting itself. Mistral's open source model is incompatible with OpenAI's closed, monetization-dependent architecture. Perplexity's answer-first approach is incompatible with Google's advertising business. ElevenLabs' developer-first free model is incompatible with how Google and Amazon build products around existing enterprise relationships. The sequence in each case was the same:
None of this required unlimited resources. It required a clear-eyed analysis of what the giant in the room was prevented from doing by their own success.
Most competitive analysis focuses on features. What does the competitor have that you don't? What can you build faster?
These three cases argue for a different question. What is the competitor prevented from doing by their own business model, customer base, or regulatory constraints? That gap is often more durable than any feature advantage because closing it would cost the incumbent more than it gains them.
Mistral's European institutional clients won't switch to a US closed model, because the constraint is structural. Perplexity's users won't go back to links, because they've stopped accepting the friction. ElevenLabs' developer ecosystem won't rebuild from scratch, because switching costs compound every time someone integrates the API into a product. Find the structural gap. Build for the buyer the giant can't reach. Then let compounding do what it does.
Why does the developer-first model work better than going enterprise-direct?
Enterprise buyers are risk-averse and slow. Developers are neither. When ElevenLabs gave developers free access to voice cloning tools in 2022, those developers embedded the product into apps, games, content tools, and workflows. By the time enterprise procurement teams started evaluating voice AI, ElevenLabs was already inside their organization through the products their internal teams were using. This bottom-up pattern is harder to compete against than a top-down sale, because switching costs accumulate at every integration point.
Can this "structural gap" approach work for early-stage founders without a big brand or network?
Yes, and it's more accessible to early-stage founders than resource-intensive strategies. A startup competing on features against a well-funded incumbent is fighting on the incumbent's terms. A startup competing on a dimension the incumbent can't address is fighting on its own terms. The analysis required isn't expensive: map what the market leader's business model requires them to do, then identify the buyers left out by those requirements. Mistral did this from a standing start in 2023 with three researchers.
What's the biggest mistake founders make when trying to apply this model?
Confusing a feature gap with a structural gap. A feature gap is something the incumbent could close in the next product cycle. A structural gap is something they can't fix without changing their business model, regulatory position, or core customer relationship. Mistral's open source positioning isn't a feature, OpenAI could theoretically open source their models, but doing so would remove their primary competitive moat. The test is simple: if the incumbent copied your move tomorrow, would it hurt them? If the honest answer is yes, the gap is structural. If not, it's just a feature, and you're in a race you'll eventually lose.
.png)
Discover how three AI startups: Mistral, Perplexity, and ElevenLabs, outsmarted giants like OpenAI, Google, and Amazon. Learn their strategies for revenue-efficient scaling.Read more
.png)
Fill out the form to get access to the tokenomics template