1. The central conclusion is that Higgsfield is much more interesting as an application and workflow company than as “another AI video-model startup.”

The company began with a strong infrastructure/foundation-model orientation. When it publicly emerged in 2024, it highlighted a custom text-to-video model and the mobile application Diffuse. By 2025–2026, however, the strategy had changed substantially. Higgsfield increasingly integrated models from multiple outside providers and built its own post-training, fine-tuning, automatic prompting, model selection, consistency systems and agentic orchestration layer on top. Reuters explicitly described it in 2026 as an application-layer company that integrates third-party models rather than simply trying to outbuild OpenAI or Google at the foundation-model level. Alex has also said that roughly 20–25 competitive models are already available on Higgsfield.

That does not make Higgsfield merely a model aggregator. Alex distinguishes aggregators, which mainly provide access to models, from Higgsfield’s attempt to control the entire workflow from intent to usable commercial output. OpenAI’s January 2026 Higgsfield case study reinforces this architecture: GPT‑4.1 and GPT‑5 are used to interpret and structure creative intent, while Sora 2 can perform the final motion generation. Higgsfield is therefore trying to own the layer between foundation models and the creative/marketing workflows historically occupied by products such as Adobe, Canva and agencies.

2. The two original founders come from very different but highly complementary backgrounds.

Alex Mashrabov brings competitive programming, Yandex search experience, a prior AI startup and successful acquisition by Snap, hands-on experience building mass-market visual AI inside Snap, Silicon Valley financing relationships, and a deep understanding of consumer distribution. AI Factory gave him an unusually complete first cycle of technology creation, productization, platform distribution and M&A.

Yerzat Dulat followed an almost opposite path: Kazakh mathematics and physics education, little conventional university education, extensive self-study in computer science and machine learning, early participation in Kazakhstan’s small ML community, open-source and research work, and then the initial launch of the project that became Higgsfield. He secured the first known angel funding and assembled a technical team before seeking an experienced cofounder who could take the project into the U.S. market. In that sense, Yerzat is best understood as the technical and project origin; Alex as the person who transformed that origin into a globally scaled company.

3. The key chronology is unusually compressed.

In 2018, Alex cofounded AI Factory with Victor Shaburov and Greg Tkachenko. The company ultimately focused on AI/computer-vision technology for faces and video and worked with Snap on Cameos. Snap acquired AI Factory in January 2020 for a reported price of approximately $166 million, after which Alex joined Snap.

In May 2023, according to Yerzat, he pitched Murat Abdrakhmanov on an infrastructure project for neural-network training and obtained $250,000. He assembled an initial team and then searched for a cofounder with greater U.S. operating experience. Nikolay “Nick” Davidov introduced him to Alex. Their scheduled roughly 30-minute conversation reportedly lasted about seven hours.

During the remainder of 2023, the concept evolved from training infrastructure toward generative media and video. Yerzat says the team moved toward video generation before OpenAI publicly unveiled Sora. Public sources generally date Higgsfield’s founding to 2023.

In April 2024, Higgsfield publicly launched with roughly $8 million in seed financing and Diffuse, a consumer/mobile product capable of generating videos from text or inserting a user from a selfie into generated footage. The initial strategy was explicitly mobile- and social-first.

The consumer mobile product grew to roughly one million users in two months, but retention proved difficult. Alex later described the discovery that serious power users were VFX professionals and marketers who needed controls that were difficult to deliver on mobile.

On March 31, 2025, Higgsfield shifted its center of gravity to a browser/desktop product. Alex regards this as the point at which the company found strong product-market fit. Click-to-Video, Draw-to-Video, camera-motion presets, identity consistency and advertising workflows followed from that change.

A $50 million Series A led by GFT Ventures followed in September 2025. In January 2026, an additional $80 million Series A extension brought total Series A financing above $130 million and the valuation above $1.3 billion.

In August 2026, Higgsfield raised a $400 million Series B at a $5.4 billion valuation. The company reported approximately $700 million in annualized revenue and more than 30 million global users, figures also reported by Reuters. Adding the roughly $8 million seed, more than $130 million across Series A financing and the $400 million Series B implies at least approximately $538 million of publicly announced major financing.

4. The decisive events were not the funding rounds but three changes in strategic thinking.

The first was Yerzat’s shift from infrastructure for training other people’s neural networks toward directly generating media. That transformed a technical infrastructure project into a potential media/creative platform.

The second was the move from mobile consumer to desktop professional. A million users did not automatically create a durable business. Professional creators and marketers showed stronger willingness to pay, clearer ROI and much deeper workflow requirements.

The third was more fundamental: moving from “we have our own video model” toward “the winning individual model matters less than the workflow controlling all of them.” Strategically, this reduces the need to compete head-on with OpenAI, Google, ByteDance, Kuaishou and other heavily capitalized model developers while turning improvements in upstream models into potential inputs for Higgsfield.

5. Alex Mashrabov’s background is better described as Central Asian/post-Soviet technical education leading into Silicon Valley than as a conventional U.S. founder story.

Public sources reveal a geographically mixed upbringing. Uzbek startup publication Pivot describes him as Andijan-born; Alex says his father is from Andijan and that both parents are professors of mechanical engineering. Forbes Russia reports that Alex and future AI Factory cofounder Greg Tkachenko studied together from the third grade at a physics-and-mathematics school in Chelyabinsk, Russia. Reducing him entirely to either a “Russian founder” or an “Uzbek founder” therefore obscures part of the story.

The most informative feature of his family background is educational rather than financial capital. Having two engineering professors as parents placed him in a strong STEM environment, while Alex himself has framed programming as a mechanism of upward mobility. The available evidence does not justify labeling his family economically wealthy.

He eventually entered the Moscow Institute of Physics and Technology. The Phystech Union identifies him as a 2018 graduate of FIVT.

His later operating style—very rapid releases, continuous experimentation and treating product-market fit as a constantly optimized problem—resembles the iterative problem-solving habits of competitive programming. That is an analytical inference from his background and operating behavior rather than a causal statement Alex has explicitly made.

6. Alex’s competitive-programming record is strong, but the frequently repeated “top three in the world” description requires precision.

He reached the ICPC World Finals twice: MIPT placed 53rd in 2014, while his Innopolis University team placed 38th in 2016. Separately, his MIPT Two Captains team placed third in CodeChef SnackDown in 2015. The accurate formulation is therefore “two-time ICPC World Finalist with a top-three result in another major international programming competition,” not “top three at the ICPC World Finals.”

A similar distinction applies to Forbes. Many short biographies call him “Forbes 30 Under 30,” but Forbes Russia’s original 2020 page describes Mashrabov and Tkachenko as nominees within its pool of 100 candidates. “Forbes Russia 30 Under 30 nominee” is the more defensible description unless stronger evidence establishes final-list status.

7. Alex’s first important professional experience was Yandex rather than Snap.

The Phystech Union records that he worked at Yandex in 2015 on Search, a fact also reported by Forbes Russia. Yandex was one of the strongest large-scale search, machine-learning and consumer-internet engineering environments in the post-Soviet technology ecosystem, giving Alex early experience connecting algorithmic work to real consumer products.

In a 2025 interview, Alex also recalled rejecting a Meta internship at age 18—historically, the company would still have been named Facebook—and later viewing that decision as important. Because the anecdote primarily comes from Alex’s own recollection, it should be treated as founder testimony rather than independently verified employment history.

8. AI Factory is essential for understanding Alex because Higgsfield is, in an important sense, his second attempt at the same broad problem.

Alex, Victor Shaburov and Greg Tkachenko founded AI Factory in 2018. Forbes Russia says the startup tested multiple computer-vision ideas before concentrating on technology capable of changing or generating appearance in video. It later collaborated with Snap on Cameos, which inserts a user’s selfie into short-form video scenes.

Snap acquired AI Factory in January 2020 for a reported roughly $166 million. Crucially, that was the company-level transaction value, not Alex’s personal cash proceeds. There is insufficient public ownership information to calculate his personal payout reliably.

After the acquisition, Alex joined Snap, continued working around Cameos, AR/AI and generative content, and eventually became head of generative AI. TechCrunch described him as Snap’s former head of generative AI when Higgsfield launched publicly in 2024.

This explains Higgsfield’s initial emphasis on selfies, short video and social sharing. The company inherited the product DNA of Cameos and Snapchat, replacing AR-style visual effects with generative video.

9. Alex later added formal business exposure to his technical background.

His public LinkedIn lists the University of California, Berkeley’s Haas School of Business for 2022–2023, although no specific degree is displayed, so it would be inappropriate to assert that he earned an MBA. His profile also records numerous Coursera certificates in machine learning, neural networks, algorithms and data analysis.

The resulting stack is unusually coherent: MIPT and competitive programming supplied algorithms; Yandex supplied internet-scale engineering; AI Factory supplied startup building and an exit; Snap supplied global consumer-product and Silicon Valley exposure; Haas added business education. Higgsfield is the first company in which all those experiences converge.

10. Yerzat Dulat followed almost the opposite path: he is fundamentally a self-taught researcher rather than a credential-driven one.

Kazakh reporting described him in 2024 as an Almaty native who was about 28 years old. He attended Kazakhstan’s Republican Physics and Mathematics School and participated in physics competitions, but did not pursue a conventional university degree. He has explained that he disliked standardized university pacing and outdated curricula and preferred self-directed study.

From roughly 2013–2015 he relied heavily on platforms such as Udacity and Coursera, studying computer science as well as mathematics, literature and philosophy. After several years focused on computer science, he discovered machine learning around 2016, when the ML community in Kazakhstan was still very small.

His career formation came through open source, remote startup projects, consulting and community activity rather than through a famous laboratory. Some secondary profiles describe him as having led ML functions at companies such as Kaspi or Kolesa, but his own detailed interview more cautiously describes project, consulting and educational work with local teams rather than clearly verifiable executive employment.

11. The famous “Yerzat rejected OpenAI” story should be sourced carefully.

Yerzat recalled that in 2018, after publishing machine-learning work, OpenAI cofounder John Schulman emailed and invited him to join the company. Yerzat says he declined because he wanted to buy GPUs and pursue his own ambitious AGI projects.

The strongest public evidence for this story remains Yerzat’s own account; there is no equivalent public confirmation from OpenAI in the sources reviewed. It is therefore most accurate to say “according to Yerzat.” Still, the anecdote is consistent with his broader pattern of choosing independent research and entrepreneurship over joining an established institution.

12. Yerzat left real research and open-source traces before Higgsfield became a commercial video platform.

In 2023, he coauthored the arXiv preprint The Poison of Alignment with Aibek Bekbayev, Sungbae Chun and James Yamazaki. The paper examined how alignment examples in supervised fine-tuning datasets can affect model performance on reasoning benchmarks, and it listed Yerzat with a Higgsfield AI affiliation.

There is also a Higgsfield open-source machine-learning project associated with Yerzat and collaborators, focused on multi-node, fault-tolerant large-model training. This supports an important historical point: the “Higgsfield” identity predates the current consumer video platform and is directly connected to Yerzat’s training-infrastructure work.

That is why Yerzat is reasonably characterized as the project’s earliest technical initiator. In May 2023, he was already pitching the infrastructure concept, raising $250,000 from Murat Abdrakhmanov and assembling the first team before Alex joined.

13. The founding structure eventually became a three-person system: Yerzat as technical origin, Alex as CEO/product/capital operator, and Mahi de Silva as strategic and institutional operator.

Yerzat says he actively searched for a cofounder who could help the company break into the United States after raising the initial $250,000. Nikolay Davidov introduced Alex, and their first conversation lasted approximately seven hours.

The company subsequently settled into an Alex-as-CEO and Yerzat-as-CTO structure. Alex has publicly said that Yerzat has served as CTO “since day one,” leading much of the technical, research and engineering side.

Mahi de Silva later became the third core founder. Alex says he had known Mahi since 2018 and that Mahi was instrumental during the AI Factory/Snap transaction. Mahi initially joined Higgsfield as an adviser, participated in major decisions and ultimately became cofounder and Chief Strategy Officer.

Structurally, the combination can be summarized as Yerzat = technical origin; Alex = product, fundraising, CEO execution and Silicon Valley positioning; Mahi = strategy, M&A experience, partnerships and enterprise relationships.

14. The first product era, Diffuse, was effectively an attempt to build the “Snapchat Camera of generative AI.”

Diffuse could generate video from text or use a selfie to place the user inside an AI-generated clip. It was intentionally mobile-first and social-first rather than a complex professional editor. When the product emerged publicly in 2024, Alex described the target market as ranging from ordinary users to professional social creators and marketers.

That concept came directly from his Snap experience. Alex later said that Snap convinced him video was the language of the internet and that a new kind of digital camera should exist for social media. Higgsfield was initially an attempt to discover that AI-native camera.

The problem was retention. A million consumer users created excitement but not necessarily recurring need. Professional marketers and VFX creators, by contrast, demanded camera control, character consistency, brand consistency, editing and bulk production. That pushed the company away from a purely mobile consumer end state.

Higgsfield also experimented with Hera, which Yerzat described in early 2025 as an AI-oriented vertical-video/streaming product being tested in the U.S. and Europe. By 2026, the company’s public product surface centered on web creation, Cinema Studio, Marketing Studio and enterprise workflows, making Hera better understood as an exploratory product than a current core asset.

15. The second product era moved from “generate a clip” to “complete a creative task.”

After the March 31, 2025 web launch, Higgsfield increasingly organized itself around controllable camera motion, presets, Draw-to-Video, Click-to-Video, image generation, identity consistency and storyboarding. By the Series A announcement, it was marketing itself as a “Video Reasoning Engine,” not merely an AI video generator.

The key design principle is to minimize prompt engineering. Higgsfield does not want marketers to spend their time learning which model requires which prompt. The platform attempts to convert a creative intention, template or cinematographic choice into the appropriate prompts, model selection and sequence of generation steps.

By 2026, the platform covered image, video, audio, editing, Cinema Studio, Marketing Studio and team collaboration. Branded/proprietary capabilities include Soul 2.0, Soul Cinema, Soul ID and Keyframes, while many final video generations still rely on external systems such as Kling, Seedance and Sora. Reuters specifically cited Soul 2.0 and Keyframes as proprietary Higgsfield products.

Its genuine technology assets therefore include workflow software, orchestration, post-training, fine-tuning, automatic prompting, model routing, proprietary Soul-family technology, behavioral feedback and enterprise relationships.

Sora, Kling, Seedance, Veo and other upstream foundation models are not Higgsfield-owned assets. Higgsfield integrates or partners around them.

16. The third era is a move from creative software toward an AI-native marketing-production system.

Alex has described the long-term workflow as ideation → creation → collaboration, followed naturally by publishing and measurement. In other words, Higgsfield wants eventually to know what performs, not merely what can be generated.

That is the commercial importance of its “reasoning engine” and agentic layer. A company selling only isolated generations is vulnerable to compression by model providers. A platform controlling trend discovery, ideation, storyboarding, generation, editing, approval, distribution, measurement and regeneration occupies a much deeper position in a marketing organization.

The Supercomputer/agentic products launched in 2026 pushed further into automated multi-scene production. Higgsfield reported that usage of these agentic products increased 42-fold during the three months following the May 2026 Supercomputer rollout and drove more than 20 million content generations per month. These are company-reported figures.

17. The business model now has three layers.

The first is individual subscriptions. Official 2026 materials showed entry pricing around $9 per month and a higher-end Ultra plan around $129 per month, with plans including different credit allowances. Credits are consumed by generations across the platform. Pricing is changeable, so these figures should be treated as mid-2026 snapshots.

The second is Team/Scale. These are seat-based plans in which monthly credits associated with users can be pooled into a shared workspace, alongside permissions, asset management, priority queues, SSO and spending controls.

The third is Enterprise. Higgsfield’s enterprise documentation describes annual, credit-based contracts sized around seats and generation volume, with dedicated capacity, security, compliance and administration. In August 2026, Alex additionally promoted dedicated U.S. GPU deployment, SSO, account management, forward-deployed creative support and per-team credit allocation.

Higgsfield is therefore not a conventional high-margin pure SaaS company. It is a hybrid of software subscription, GPU inference consumption and enterprise contracting. Alex himself says SaaS is the closest business-model analogy, while acknowledging the very different economics of generative AI.

Model orchestration is consequently also a margin strategy. Open models can be fine-tuned and distilled for lower-cost specialized workflows, while closed models can be obtained through provider relationships. Alex explicitly links open-source models to the ability to improve gross margins on certain use cases.

18. Higgsfield’s capital network became global almost immediately.

The first identifiable financing came from Murat Abdrakhmanov, who according to Yerzat invested $250,000 in May 2023 and enabled the initial team to form before Alex joined.

The approximately $8 million public seed round in 2024 included Menlo Ventures as a central investor alongside names including Charge Ventures, BITKRAFT, K5 Tokyo Black, AI Capital Partners and DVC. Menlo’s backing gave the company an early connection to top-tier U.S. venture capital.

The $50 million Series A in 2025 was led by GFT Ventures with participation from BroadLight Capital, NextEquity Partners, AI Capital Partners, Menlo Ventures and Alpha Square Group, among others. GFT’s Jeff Herbst became an important board and capital relationship.

The January 2026 $80 million Series A extension included Accel, GFT and Menlo and pushed the valuation above $1.3 billion.

The August 2026 $400 million Series B was of an entirely different scale. DST Global led, with new participation from Tribe Capital, Growth Equity at Goldman Sachs Alternatives, Smash Capital, Fifth Wall, Valor Capital, Intel Capital, Liberty Global Tech Ventures, Mirae Asset Capital and NTT DOCOMO Ventures. Accel, Menlo, AI Capital Partners, GFT, Capra Ventures, BAM Corner Point and BroadLight returned. Natalia Vodianova Arnault joined as both investor and adviser.

That portfolio now spans not just AI venture funds but compute, finance, real estate, media, telecommunications, fashion and distribution—strategic relationships that may provide industry access as well as capital.

19. Relationships created during the AI Factory era were directly reused at Higgsfield.

Mahi de Silva is the clearest example. Alex says Mahi had been close to him since 2018, helped navigate AI Factory’s $166 million sale to Snapchat, then entered Higgsfield as an adviser before becoming the third cofounder.

AI Capital Partners/Alpha Intelligence Capital is another durable relationship. An investor associated with the firm has said the group wrote an early Higgsfield check in December 2023, invested in every subsequent round and later placed Antoine Blondeau on the board.

Meanwhile, the Yerzat–Murat Abdrakhmanov–Nikolay Davidov network represents the Central Asian/Eastern European side of the company’s origins. Higgsfield is therefore best understood as a junction among Central Asian technical talent, post-Soviet engineering/entrepreneurship and U.S. venture/Big Tech networks.

20. Its assets should be divided into economic assets and influence assets.

The first economic asset is equity in the private company Higgsfield. Its latest financing valuation is $5.4 billion, but a private-round valuation is not a public-market capitalization and certainly not immediately liquid founder wealth.

For Alex, Forbes Russia cited expert estimates that his post-Series-B ownership could still be approximately 20%–30%. A Higgsfield representative disputed the estimate but did not disclose the actual percentage, citing confidentiality. Alex’s true ownership percentage and precise net worth therefore remain unconfirmed. Multiplying an assumed stake by the $5.4 billion valuation should not be presented as established personal wealth.

Yerzat’s exact ownership is likewise not publicly established through a reliable cap table. His early-founder role makes equity an obvious core economic interest, but the percentage cannot be responsibly inferred.

A second asset category is software and IP: Soul-family technology, post-training, routing, orchestration, workflows, enterprise infrastructure and accumulated user feedback. Reuters specifically identified Soul 2.0 and Keyframes as proprietary Higgsfield tools.

A third category is influence: more than 30 million users, creator communities, Academy content, presets and templates, filmmaking projects, agencies, brands and studio relationships. These may not appear directly on a balance sheet, but they help prevent Higgsfield from becoming a replaceable front end for upstream models.

21. Higgsfield’s most impressive achievement is the speed with which it converted “AI-video demos” into paid commercial workflows.

Reuters reported in January 2026 that social-media marketers represented roughly 85% of Higgsfield usage. While much of the industry was debating whether AI could make feature films, Higgsfield had located a more immediate market: organizations that must produce large quantities of TikTok, Instagram, YouTube and advertising content every day.

Alex has given an aggressive customer-economics example in which a minute of conventional commercial production might cost around $100,000 while an AI-first workflow could in some situations fall to roughly $500. That is Alex’s estimate and should not be treated as a universal industry cost benchmark. Its real significance is the sales argument: the same marketing budget can test dramatically more creative variants.

By August 2026, Reuters reported approximately $700 million in annualized revenue and more than 30 million users, roughly double the user base from January. Higgsfield also says it serves 390 Fortune 500 companies.

The accounting language matters. Reuters corrected its January 2026 article to clarify that the then-$200 million figure was an annualized run-rate projection, not recognized revenue. The later $300 million, $500 million and $700 million figures should therefore be read as annualized operating rates rather than audited revenue already earned during a complete fiscal year.

22. Yerzat’s impact in Kazakhstan already extends beyond his formal CTO position.

On January 28, 2026, President Kassym-Jomart Tokayev met Yerzat Dulat and investor Murat Abdrakhmanov to discuss AI, venture capital and technology entrepreneurship. Kazakhstan’s presidential office explicitly framed Higgsfield as an example of the kind of global AI champion the country wants to produce.

At the meeting, Yerzat stated that nearly 95% of the Higgsfield team consisted of Kazakh specialists and expressed willingness to cooperate on AI talent development. The 95% figure is Yerzat’s statement, not independently audited workforce data, but it demonstrates Higgsfield’s symbolic importance: a company can draw heavily on Central Asian engineering while operating with global capital, customers and headquarters.

Alex has similarly described Almaty as a strategic engineering hub and emphasized Central Asia’s competitive-math, programming and physics talent. Higgsfield therefore has an unusual organizational structure: California as headquarters and capital center, with a substantial portion of its engineering culture rooted in Central Asia.

23. The most serious public controversy concerns Higgsfield’s growth and marketing practices in early 2026 rather than earlier personal scandals involving the founders.

A February 2026 Forbes investigation found that some Vibe Motion marketing materials supplied by Higgsfield to creators were not AI-generated outputs but appeared to be existing Envato stock/video templates carrying Higgsfield branding. Mahi de Silva acknowledged that the material had been prepared for internal ideation and was mistakenly distributed externally, saying company processes had broken down.

More seriously, Forbes reviewed promotional material containing racist or obscene depictions of well-known children’s characters and nonconsensual AI deepfakes involving public figures including Sydney Sweeney, Zendaya and Donald Trump. Mahi confirmed that material had been created by both internal marketing personnel and third-party creators and distributed externally, and said it was a mistake. Alex later acknowledged that rapid scaling had caused internal processes and external communications to fall behind the company’s stated values. Higgsfield said future external materials would require mandatory legal and senior-leadership review.

The issue is especially significant because Higgsfield’s own policies prohibit unauthorized use of likenesses. It therefore exposed a central governance conflict in generative media: the incentives for viral, provocative marketing can directly collide with likeness, copyright and safety policies.

Forbes also reported that Higgsfield’s official X account was temporarily suspended for what Mahi said the platform described as “inauthentic behavior.”

24. A second set of controversies concerns growth quality: heavy discounts, “unlimited” plans, compute congestion and creator payments.

Some users told Forbes that after purchasing heavily discounted unlimited plans, generation performance became severely throttled after limited usage. Mahi acknowledged periods of service degradation caused by heavy traffic and bot attacks. He said approximately 40,000 suspected automated accounts had been shut down and that the company had refunded approximately $1.35 million to legitimate users affected by service problems.

The Higgsfield Earn creator-incentive program also generated complaints about delayed withdrawals, disappearing submissions and unexplained account bans. Higgsfield attributed part of the problem to fraud, including content not created with Higgsfield and artificially inflated engagement, and said approximately 90% of submissions had been paid.

None of this proves that Higgsfield’s business model is unsustainable. It does, however, make customer-quality analysis important: how much growth comes from long-retention professional customers, and how much from discounts, free credits, influencer incentives and short-term viral acquisition? Forbes quoted investors expressing similar questions about the economic flywheel.

25. A third controversy is cultural: Higgsfield has deliberately adopted an aggressive, provocative growth posture.

Mahi acknowledged to Forbes that the company studies what receives attention on networks such as X and knows controversial material often travels farther. Higgsfield also faced backlash after marketing one automation feature with language claiming that it had effectively eliminated 20 creative jobs.

That posture exposes a tension in Alex and Yerzat’s technological optimism. Higgsfield’s economic proposition is precisely that AI can reduce the cost of filming, VFX and advertising production, so it inevitably intersects with creative labor displacement. Higgsfield frames the change as reducing “production tax” so that better ideas can win; critics can reasonably describe the same productivity shift as commoditizing skilled creative work.

26. As of August 23, 2026, Alex Mashrabov is best understood as one of the most aggressive commercializers of the generative-visual application layer.

He remains Higgsfield’s cofounder and CEO. The company’s latest private valuation is $5.4 billion; its investor base includes DST Global, Goldman Sachs Alternatives, Intel Capital, Accel, Menlo and GFT; and the platform reports more than 30 million global users.

His rare advantage is not simply understanding AI models. He has now built twice around the same broad pattern: face/visual AI → consumer or commercial content product. AI Factory/Cameos was acquired by Snap; Higgsfield is attempting to turn the same underlying product instinct into a broad visual-production platform.

Alex is therefore better characterized as a product/market founder with deep technical literacy than as a foundation-model scientist. His primary skill is recognizing, when model capability becomes sufficient, which interface, distribution channel, user group and workflow can convert that capability into revenue fastest.

27. Yerzat Dulat occupies a different real-world position: he has become a model of Central Asian technical entrepreneurship.

His credibility was built through competitions, self-education, open source, research and entrepreneurship rather than elite American degrees, famous laboratories or Big Tech titles. He also appears to have initiated the earliest version of the Higgsfield project in 2023.

Today he is Higgsfield’s cofounder and CTO, with value extending beyond model training to building a global technical organization around Kazakh engineering talent. His formal meeting with Kazakhstan’s president demonstrates that his symbolic importance has already entered the country’s national technology-development narrative.

28. The founders’ division of labor can be summarized simply: Yerzat recognized the technical window; Alex recognized the commercial form.

Yerzat saw an opportunity in AI infrastructure and generative media in 2023, secured initial capital and assembled a technical team. Alex brought Snap-era video-product experience, U.S. investor access, consumer-internet intuition and eventually pushed the target from a single generative model toward an AI-native creative operating layer.

Mahi de Silva later supplied a third element: seasoned Silicon Valley strategy, M&A, partnerships and enterprise experience. The combination helps explain how a project rooted heavily in Kazakhstan could rapidly become a U.S.-headquartered private company valued at $5.4 billion.

29. Higgsfield’s strongest moat today is not any single model.

Its defensibility comes from the combination of professional visual workflows, multi-model routing and orchestration, proprietary post-training and consistency technology, extremely high product-release velocity, creator/marketing feedback data, and enterprise/customer relationships. Alex says the company releases products roughly six days a week, more than 300 times per year, using faster application-layer iteration to exploit the slower development cycles of large foundation models.

Its greatest structural risk is the mirror image of that advantage. Upstream model providers are adding editing, agents and enterprise tooling themselves; inference remains expensive; Higgsfield does not control the major external foundation models; and customers can simultaneously use Adobe, Canva, OpenAI, Google, Runway, Kling and others. Higgsfield must continuously prove that its workflow layer creates substantially more value than simply going directly to the model provider. This is an inference from the company’s current multi-model architecture.

30. The best final description of Higgsfield is that it is competing for the application-control layer of the generative-visual era.

If foundation models become increasingly commoditized, users will not want to decide each day which model is best for a product ad, which one produces the best cinematic shot, or which one preserves character identity. The platform that makes those decisions automatically, retains brand assets, coordinates team approval, generates the content and ultimately measures performance could become a new creative-software gateway. Higgsfield’s strategy is built around that thesis.

If that thesis succeeds, the $5.4 billion valuation is better interpreted as a bet on a potential AI-native Adobe/Canva/creative-operations platform than as a valuation of a simple AI video generator. If foundation-model companies vertically integrate workflow, editing, consistency and enterprise management, however, the middle layer occupied by Higgsfield could be heavily compressed. That is the company’s central strategic fork.

At the founder level, Alex’s defining achievement is having twice converted frontier visual AI into large-scale consumer or commercial products. Yerzat’s defining achievement is demonstrating that a self-taught researcher from Kazakhstan, outside the conventional elite-university path, could initiate a technical project and use a cross-border founder team and global capital network to scale it internationally. Higgsfield itself is one of the clearest current examples of the combination Central Asian engineering + Silicon Valley capital + global foundation models + creator-economy distribution.