Startup Failures 2025: The Complete Retrospective Report
The 2025 thesis broke into three storylines: AI wrappers without moats unraveled fast, capital-intensive moonshots ran out of bridge financing, and enterprise SaaS started feeling the first real churn from AI-native alternatives. Looking back, 2025 was the inflection point — not the bottom, but the moment the market stopped pretending hype could substitute for unit economics.
- 343tracked shutdowns
- $302.4Bcapital destroyed
- 8sectors hit
- 7.5×lower risk validated
343 tracked collapses analyzed in this report.
What's actually driving 2025 failures
The AI hype premium collapsed
Series A/B AI startups raised at 80-120x ARR in 2024. By Q3 2025, the median had compressed to 18x as growth missed plan and gross margins came in 30-50% below SaaS comparables.
Governance failures became existential
Builder.ai's collapse wasn't about product — it was about a $1.5B house of cards built on round-tripped revenue and offshore engineering theater. Diligence standards reset hard after this one.
Capital-intensive moonshots ran out of road
Lilium, Volocopter, Northvolt and others proved that even $1B+ raised can't outrun a broken unit economic model. Hardware moonshots without a clear path to gross margin positivity were the year's biggest dollar losses.
Vertical farming, micromobility, and Q-commerce died quietly
Three of the 2021 darling categories were effectively dead by year-end. Plenty, Bowery, and several Q-commerce players showed that beautiful decks cannot fix negative contribution margin.
343
notable failures tracked
$302.4B
total capital destroyed
8
industries affected
Post-Pandemic Demand C…
top root cause
2025 failures — sliced by sector and cause
Failures by Industry
Failure Reasons
The 2025 Timeline
Live tracking of every major shutdown wave as it unfolds.
The AI-wrapper shakeout begins
Dozens of seed-stage AI wrappers quietly wound down as OpenAI shipped competing native features. The first credible signal that "thin layer over GPT" was not a venture-backable business.
Builder.ai implodes — governance shock
Builder.ai collapsed in May after revenue restatements and a $50M cash sweep by lenders. The largest AI-adjacent failure since the ZIRP era and a wake-up call for diligence on revenue quality.
Hardware moonshots run out of bridge financing
Lilium filed for insolvency for the second and final time. Northvolt entered Chapter 11. Several climate-hardware Series B companies hit wall after wall trying to raise extension rounds.
Vertical farming and Q-commerce close out the year
Plenty filed Chapter 11. Bowery wound down operations. Q-commerce consolidation accelerated as Getir's retreat became permanent. The 2021 thesis basket was officially closed.
Sectors to Watch in 2025
Where the next wave of shutdowns is most likely to land.
AI wrappers (horizontal)
No moat. Native OpenAI/Anthropic competition. Gross margins compressed below 25% for most.
Electric aviation / eVTOL
Lilium and Volocopter proved certification timelines outrun any reasonable funding plan.
Vertical farming
Unit economics fundamentally broken at scale. Plenty and Bowery confirmed the thesis is dead.
EV manufacturing (non-Tesla, non-BYD)
Fisker, Canoo, Arrival, Lordstown — the entire 2020-2021 SPAC cohort is effectively gone.
Quick commerce / 15-min delivery
Getir's retreat marked the end. Negative contribution margin per order never closed.
Healthcare AI (consumer-facing)
Regulatory pressure rising. Reimbursement pathways unclear for most direct-to-consumer plays.
Key Highlights of 2025
Builder.ai's $1.5B collapse became the largest governance-driven failure since Theranos
Lilium's second and final bankruptcy ended the 2020 eVTOL hype cycle definitively
Northvolt filed Chapter 11 despite $15B+ raised — Europe's largest battery bet failed
Plenty's Chapter 11 confirmed vertical farming is structurally uneconomic at scale
AI-wrapper shutdowns hit triple digits as OpenAI shipped native competing features
Median Series B AI valuation compressed from 80x ARR (2024) to 18x ARR (Q4 2025)
Biggest Failures of 2025
Click any card for the full post-mortem.
Byju's
Aggressive acquisition-driven growth funded by debt is fragile. Transparency with investors is non-negotiable.
Didi (DiDi Global)
Going public in the US against your home government's wishes can trigger an existential regulatory response that no amount of funding can overcome.
Juul Labs
A $38B e-cigarette company that hooked teenagers on nicotine faced total regulatory destruction.
Northvolt
Manufacturing battery cells at scale is extraordinarily hard. Even $13.8B couldn't bridge the gap between lab results and factory output.
Grab Holdings
Building a super-app across fragmented Southeast Asian markets with ride-hailing, delivery, and fintech requires massive capital and patience — profitability may take a decade or more.
Rivian (Value Destruction)
Rivian IPO'd at $150B — briefly worth more than Ford and GM. The stock fell 90% as production couldn't match hype.
C3.ai
C3.ai is what happens when enterprise AI promises outpace enterprise adoption. Despite a billionaire founder and a $10B IPO valuation, revenue barely grew while the stock lost 85%+.
Fastly
Fastly was the 'developer-friendly CDN' that rode TikTok's growth to a $10B market cap. When TikTok optimized its CDN spend, Fastly lost its largest customer and 85% of its stock value.
DataRobot
AutoML was a brilliant concept when data science was scarce. But as AI tools became ubiquitous and cloud providers offered their own AutoML, DataRobot's $6.3B valuation evaporated.
23andMe
A one-time-purchase consumer business cannot sustain a public-company cost structure. A brand cannot survive losing the data.
Predictions for 2026
2026 will see the first major foundation model company face insolvency or fire-sale
Legacy B2B SaaS churn from AI-native challengers will hit a 25-year high
Vertical AI with proprietary data will survive — generic horizontal AI will not
Crypto/Web3 zombies from 2021 will formally dissolve as cash reserves expire
D2C consumer hardware brands face a 25%+ failure rate as CAC inflation persists
Learn from Startup Failures
93% of startups fail. Study these cases to avoid the same mistakes.
Silicon Valley Bank (SVB)
Concentration risk in a single industry can create correlated failure modes, making a business vulnerable to market shifts.
BitMEX
You cannot build a financial empire by deliberately evading regulations. BitMEX's founders chose offshore structures over compliance and paid with criminal convictions.
Terraform Labs (Terra/Luna)
Algorithmic stablecoins backed by their own volatile sister token are reflexive ponzis waiting to unwind. Yield that high implies risk that high.
WeWork
Valuation hype cannot mask fundamentally broken unit economics. Corporate governance failures amplify founder risk.
Frequently Asked Questions
How many startups failed in 2025?
Public databases including IdeaProof's corpus track several hundred notable shutdowns per year — but the real number including unannounced wind-downs is in the tens of thousands. 2025 is on pace to set a post-2001 record for B2B SaaS shutdowns specifically.
What are the biggest startup failures of 2025?
The biggest failures are dominated by AI wrappers without data moats, capital-intensive moonshots that ran out of bridge financing, and legacy B2B SaaS losing to AI-native challengers. See the 'Biggest Failures' section above for the live list.
Why are so many AI startups failing in 2026?
Three reasons converge: (1) gross margin compression as inference costs stay high relative to seat-based pricing, (2) feature parity with native OpenAI/Anthropic tools eliminates the wedge, and (3) lack of proprietary data or workflow lock-in means zero defensibility.
How do I avoid being on this list next year?
Validate demand before building, prove unit economics work at small scale, identify a defensible moat (data, distribution, or workflow lock-in), and stay default-alive on at least 18 months of runway. IdeaProof's AI validator runs this exact analysis in ~120 seconds.
How often is this report updated?
The 2025 report is updated quarterly with new shutdowns, sector heatmaps, and forward-looking risk shifts. Aggregate statistics recompute on every deploy from the live IdeaProof failure database.