The Twenty Minute VCVictor Riparbelli, CEO @Synthesia: OpenAI vs Anthropic vs X.ai - Who Wins and Why | E1246
CHAPTERS
- 0:00 – 1:31
Series D announcement: $100M round and the “escape velocity” ambition
Victor opens with the big news: Synthesia’s $100M Series D led by NEA, framing it as fuel to reach “escape velocity” and win the category. The conversation sets up a recurring theme: building a real business with durable unit economics while still playing to win big.
- •Synthesia raises $100M Series D led by NEA
- •Goal: build a war chest to “shut down the category”
- •Emphasis on scaling responsibly while pursuing a massive outcome
- •Early framing of competitiveness and category leadership
- 1:31 – 2:48
The seed round in 2017: pitching generative video before the market was ready
Victor recounts Synthesia’s early fundraising when “generative AI” wasn’t yet mainstream and investors largely didn’t understand the vision. The team believed GANs would transform content creation, but the pitch felt too futuristic—especially in Europe.
- •Early thesis: generative AI would reinvent content creation, starting with video
- •Market context: AI was mostly seen as data analysis/decisioning, not creation
- •Investors rejected the vision as too far-out; Europe especially skeptical
- •The long-term pitch: Hollywood-quality creation from a laptop
- 2:48 – 4:42
Why Europe said no: PE mindset, not technologist conviction
Victor explains why fundraising was so hard in Europe: many investors were steeped in finance and wanted Excel-model clarity rather than technology leaps. After 80–90 rejections, a small seed came together—highlighting how narrative risk was priced then.
- •European VC skewed toward private equity-style evaluation
- •Deep tech visions are hard to underwrite via spreadsheets
- •Dozens of rejections before finding a believer
- •Seed terms: $1M raised at a $5M post-money; Mark Cuban as key backer
- 4:42 – 6:03
Early constraints as an advantage: focus, revenue discipline, and avoiding distractions
The discussion turns to whether bigger early rounds would have helped. Victor argues capital scarcity forced focus on customers and revenue from day one, preventing the team from chasing distractions like deepfake detection that the market wanted at the time.
- •Belief: raising more early would have reduced focus
- •Temptation then: build deepfake detection because the world equated AI video with deepfakes
- •Constraints drove customer obsession and early monetization (charging from day one)
- •More money increases options—and the risk of losing discipline
- 6:03 – 7:34
Too much money too early: why you can’t buy product-market fit
As an angel investor, Victor critiques oversized “pedigreed founder” rounds. He stresses that PMF requires time and learning, and that hiring big teams pre-PMF often slows iteration and pushes founders to outsource core learning to PMs/sales too soon.
- •“Too much money too early is not healthy”
- •PMF is time-bound learning; money can’t shortcut it
- •Large pre-PMF teams create coordination drag and false confidence
- •Founders shouldn’t hire PM/sales too early; PMF discovery is the founder’s job
- 7:34 – 8:44
PMF is never finished: stacking product-market fits as the company expands
Harry challenges the static notion of PMF; Victor agrees and reframes success as a series of PMFs across segments and products. The founder’s enduring job is to push toward the next PMF while teams scale what already works.
- •PMF evolves across personas (creators → SMB → enterprise)
- •A successful company stacks multiple PMFs over time
- •Teams can scale proven PMFs; founders should pursue the next frontier
- •Strategic focus: where the company must be in 2–3 years
- 8:44 – 13:01
How Synthesia becomes a $50–$100B company: video replaces text as default communication
Victor outlines a sweeping thesis: text is scalable but low-fidelity, and AI will make video/audio creation as scalable as text. Synthesia’s true market isn’t “video production” but the vast world of text and slides that can be converted into video workflows.
- •Communication shift: from text-first to video/audio-first consumption
- •AI removes dependency on cameras/mics, making video scalable like text
- •Future: interactive video for sales, support, and product understanding
- •Market definition: target ‘text + slides’ and convert a slice into video
- 13:01 – 16:02
Raising as strategy: deterring competitors vs funding product excellence
Harry asks if capital is being used as a competitive weapon. Victor acknowledges signaling effects but insists Synthesia doesn’t run based on competitors; instead, it raises to build the best product and stay opportunistic with a strong balance sheet.
- •Capital can backfire if you don’t know how to deploy it
- •Raising does send a market signal, but shouldn’t drive decisions
- •Competitors provide useful feedback loops once a category forms
- •Category marketing by competitors can lift the leader with the best product
- 16:02 – 19:44
Where we are in the AI hype cycle: enterprise pilots, unclear demand, and the renewal trap
Victor diagnoses enterprise AI demand as exploratory: buyers are told to ‘have an AI strategy’ but don’t know what they need. This creates a dangerous illusion for startups optimizing for pilots and new contracts rather than real value proven through renewals.
- •Enterprise buyers have budget but often lack clarity on actual needs
- •Pilots/POCs are easy to sell; ROI proof is harder without deep customer insight
- •The ‘wall of churn’ arrives when 12-month contracts don’t renew
- •Key metric: renewals, not just closing new contracts
- 19:44 – 22:44
What’s getting funded that won’t last: buzzwords, “AI employees,” and demo-driven thinking
Victor flags hype-patterns: obsession with buzzwords (like “agents”) and anthropomorphizing software as “AI employees.” He argues real adoption comes from useful products, not narratives that overpromise autonomy or misframe what software is.
- •Buzzword-led pitches trigger skepticism (especially “agents” used vaguely)
- •Calling tools ‘AI employees’ is misleading and unhelpful for adoption
- •Tech industry often sets expectations it can’t meet on timelines
- •Demos are not production-ready transformation at scale
- 22:44 – 25:32
Commoditization of foundation models: distribution and workflow products win
Victor agrees text generation is commoditizing and shifts the lens to product scaffolding and distribution. He also gives his take on OpenAI vs Anthropic vs xAI, emphasizing asymmetric upside and distribution/data advantages through owning X.
- •Text-generation layer becoming ‘good enough’ for many use cases
- •Competitive edge shifts to product scaffolding, pricing, and distribution
- •Investment pick: xAI for asymmetric upside and X distribution/real-time data
- •Compute matters, but algorithms and efficiency breakthroughs can reshape the race
- 25:32 – 30:29
The real differentiator: control, safety/moderation, and workflow—not just model capability
Victor argues the next breakthroughs aren’t only fidelity but controllability—being able to reliably direct outputs (especially in video). The conversation also touches moderation strategy and why Synthesia, as an enterprise product, chooses stricter controls aligned with customer expectations.
- •Key challenge: adding layers of control without degrading output quality
- •Moderation is shifting (community notes-style approaches vs human review)
- •Synthesia’s stance: enterprise brand safety over maximal ‘free speech’ posture
- •Hardest problem is the ‘gray zone’ content, not obvious green/red cases
- 30:29 – 37:31
Will model providers move into apps? Synthesia’s moat as a full video workflow platform
Addressing fears that OpenAI will ‘move into this,’ Victor reframes Synthesia as a workflow company, not an avatar company. Customers buy an end-to-end value chain: ideation → editing → multilingual distribution → hosting/publishing, with models as one component.
- •Misconception: Synthesia is ‘just avatars’; reality: customers buy workflow
- •Enterprise value: efficient message delivery, engagement, multilingual delivery
- •Strategy: own a narrow model domain (humans presenting) tied to a broader platform
- •Model mix: uses OpenAI most, with some workloads shifting to Anthropic/Gemini; pricing matters
- 37:31 – 48:12
Future of content creation: creation cost goes to zero, discovery becomes the battleground
Victor predicts true democratization: content can be generated digitally without physical capture, pushing creation costs toward zero and flooding supply. The key limiter becomes discovery and taste—what gets surfaced—and he points to TikTok’s interest graph as a powerful model.
- •Shift from physical capture to digital generation across video/audio
- •Creation cost trends toward zero → massive increase in content volume
- •Discovery/surfacing determines what matters more than raw supply
- •TikTok’s interest-based graph and rapid testing loops as a model for discovery
- 48:12 – 53:54
Identity, provenance, and cross-platform trust: ‘Shazam for content’ and verification by default
Victor outlines a provenance-first future: verify who created content, when, and how, down to the asset level. He imagines content fingerprinting akin to YouTube’s copyright detection—potentially using a shared registry—to make unverified content the anomaly and reduce fraud/disinformation.
- •Need for identity verification in major platforms and at the content level
- •Concept: universal content fingerprinting (‘Shazam for everything’)
- •Flip the default: verified content is normal; unverified should stand out
- •Use cases: tracing edits, detecting reused war footage, linking clips to originals
- 53:54 – 56:13
Labor market impact: technical roles fade, creativity and ‘human + AI’ workflows expand
Victor expects some technical production roles to diminish while creative and editorial judgment becomes even more valuable. He argues most jobs will transition rather than vanish, with AI accelerating first-pass work and humans providing taste, direction, and final decisions.
- •Some production tasks (e.g., camera operation) reduce over time
- •‘You + AI’ beats either alone: faster first pass, human judgment for quality
- •More creators overall, similar to how text creation became universal
- •Adoption rate limiter is people: trust, integration, and behavior change
- 56:13 – 1:05:41
Building Synthesia in London vs the US: talent, loyalty, policy, and the UK’s startup brand
Victor explains he started in London partly due to immigration constraints but discusses the trade-offs candidly. He sees the US as higher-probability for success, yet argues London/UK has major talent advantages and can win by protecting founder incentives, avoiding overregulation, and improving the country’s ‘brand’ for builders.
- •US advantage: ecosystem density; UK catching up with global leaders
- •London advantages: talent cost/availability and (often) higher loyalty
- •Policy priorities: keep entrepreneur reliefs, consider GPU/data-center support, avoid EU-style overregulation
- •UK brand matters: signal ‘open for business’ and improve fundamentals; LSE liquidity remains a challenge
- 1:05:41 – 1:13:29
Quick-fire: games as training, generalist advantage, secondaries, and investor loyalty
In rapid Q&A, Victor shares personal beliefs: strategy games build decision-making through fast feedback loops, and being a curious generalist helps pattern-matching and leadership. He also reflects on taking secondaries to reduce stress and credits Mark Cuban for believing early when others didn’t.
- •Computer games (e.g., strategy/MMOs) as decision-making simulators
- •A ‘generalist’ mindset: diverse inputs improve analogies and choices
- •Regret: not doing a pure STEM degree; business can be learned on the job
- •Secondaries can free founders to make better long-term decisions; would pick Mark Cuban again
