Aidan Gomez

CEO of Cohere and co-author of the 'Attention is All You Need' paper

Mentions

Appeared in

Discussed in

Key positions and views

The 'scale is all you need' hypothesis is breaking down due to diminishing returns; future AI progress will come from new capabilities like learning from experience, not just bigger models [48, 56, 16, 55].

A vertically integrated strategy, where a company builds both the foundation model and the end-user application, is crucial for superior product quality in the enterprise space [3, 20].

The AI market is bifurcating, with Cohere focused on the enterprise, while competitors like OpenAI, Google, and Meta target the consumer market [38, 42, 65].

The most significant near-term AI risk is geopolitical, stemming from state-level bad actors gaining advanced capabilities, making it imperative for liberal democracies to maintain a lead [47].

For enterprise adoption, model efficiency and deployment flexibility (cloud-agnostic, on-premise) are more important than raw model size, leading Cohere to strategically cap its models to run on a maximum of two GPUs [11, 35, 52].

Podcast consensus on Gomez

Points of consensus

▶Gomez consistently argues across multiple sources that the 'scale is all you need' approach to improving AI models is hitting diminishing returns and starting to saturate [48, 56].Apr 2026

▶He repeatedly predicts that the next major breakthrough in AI will be the ability for models to learn from experience and user interactions, a feature he expects in the next model generation [16, 55, 63].Apr 2026

▶He believes that if his team at Google had not invented the Transformer, another research group would have independently developed a similar architecture within 12 to 18 months [6, 53].Apr 2026

▶Gomez projects that the scale of AI training will increase dramatically, from the current tens or hundreds of thousands of GPUs to millions of GPUs within a couple of years [49, 66].

Points of debate

▶Gomez's claim that the top 5-7 large language models have converged in capability and are now 'interchangeable' [59] would be strongly contested by competitors who market their models based on unique performance advantages.Apr 2026

▶His assertion that AI will augment workers rather than cause mass unemployment due to 'infinite demand' [39] represents a classic economic viewpoint that is heavily debated against fears of technological displacement.

▶Gomez's dismissal of AI self-awareness claims as 'BS' [12] and his belief that doomsday rhetoric was a strategic posture by competitors [57] places him in direct opposition to factions within the AI community that treat existential risk and consciousness as serious, pressing concerns.Apr 2026

▶He claims Google has technologically surpassed OpenAI in model capability [13], a contentious point in the highly competitive AI landscape where performance leadership is constantly debated and benchmarked.Apr 2026

Key themes

▶The Enterprise AI PlaybookApr 2026

Gomez outlines a clear strategy for winning the enterprise market, emphasizing cloud-agnostic deployment, on-premise capabilities, and model efficiency over raw scale. This approach, exemplified by Cohere's 'two GPU constraint' [52], contrasts with the consumer-focused strategies of competitors like OpenAI [42].

This focus on practical enterprise needs (cost, security, integration) over chasing the largest possible model could give Cohere a durable advantage in the B2B space, even if its models do not top consumer-facing leaderboards.

▶Beyond Scale: The Next AI FrontierApr 2026

Gomez consistently argues that the era of easy gains from scaling models is ending, citing diminishing returns [48, 56]. He posits that the next breakthroughs will involve new capabilities like learning from experience [16, 63] and the strategic use of synthetic data, which he claims now constitutes the majority of training data at all major labs [50].

Investors should watch for R&D that moves beyond simple scaling laws and focuses on novel architectures and learning paradigms, as this is where Gomez believes future value and differentiation will be created.

▶A Pragmatic View of AI Risk and Geopolitics

Gomez dismisses existential AI risk narratives as overblown and potentially strategic posturing [57, 64]. His focus is on tangible, near-term risks, particularly the geopolitical implications of state actors gaining access to advanced AI and the need for liberal democracies to maintain a strategic advantage [47].

Gomez's geopolitical framing of AI competition suggests that national security and government contracts will be a critical battleground, potentially favoring companies like Cohere that can offer secure, on-premise deployments [11].

▶The Consolidation of the AI Stack

Gomez predicts a market evolution from a fragmented landscape of best-of-breed point solutions to consolidated, integrated platforms [33]. He believes enterprises will ultimately prefer a handful of powerful, versatile models over managing hundreds of specialized ones for each department [25].

This prediction suggests a future with a few dominant platform players. Companies offering niche AI solutions may face acquisition or obsolescence, while those building integrated platforms, like Cohere with its North application layer [20], are positioned for long-term market leadership.

Source episodes

Sentiment over time

Not enough data for timeline

Changes over time

2017

Co-authored the 'Attention is All You Need' paper at Google over a 12-16 week period, introducing the Transformer architecture [7]. Gomez believes Google did not fully grasp the paper's significance at the time [68].

Post-2017

Founded Cohere in Toronto, Canada, at a time when he estimates only 15-30 people in the world had experience training large language models [1]. He considers the Toronto location a key strategic advantage for talent acquisition [41].

c. 2023-2024

Articulates that the AI industry is hitting diminishing returns on scaling [48, 56] and that the top models have converged in capability [59]. During this period, he claims synthetic data became the dominant source for training new models across all major labs [50].

Present Day

Leads Cohere as it experiences high demand, necessitating a rapid expansion of its sales and engineering teams [2]. The company's focus is on deploying a narrow set of use cases at massive scale within enterprises, often to tens of thousands of employees [51].

Near Future

Predicts the next generation of models (e.g., GPT-6) will feature the ability to learn from experience [55] and that the hardware scale for training will reach millions of GPUs within a couple of years [49].

Suggested prompts

How does Aidan Gomez's 'two GPU constraint' philosophy for model development reflect his broader thesis on the limits of scaling and the specific needs of the enterprise market? &nearr;Gomez claims the top LLMs have 'converged,' yet he also details specific strategic differences between Cohere, OpenAI, and Anthropic. How can both of these views be simultaneously true from his perspective? &nearr;What are the second-order effects for the AI industry if Gomez is correct that synthetic data now constitutes the majority of training data for all major labs? &nearr;Gomez frames AI risk primarily as a geopolitical concern between state actors. How does this perspective shape Cohere's product strategy, particularly its emphasis on on-premise and private cloud deployments? &nearr;

Key concepts

Enterprise AI Strategy 2 ep AI Model Development & Scaling 3 ep AI Industry Competition 3 ep Transformer Architecture 3 ep Synthetic Data 2 ep Cohere's Business Strategy 2 ep AI Risk & Geopolitics 2 ep Future of AI 2 ep AI Agents 1 ep

Notable quotes

“the whole like scale is all you need hypothesis is breaking. We're very heavy into diminishing returns of capital and compute.”

Aidan Gomez · Aidan Gomez: Scaling Limits Emerging, AI Use-cases with PMF & Life After Transformers

“we set a constraint saying within two GPUs, we are going to squeeze the most intelligence out of those two GPUs that we can. But that's the cap. If the model fits on three GPUs, we're not building that model. It's two GPUs.”

Aidan Gomez · Synthetic Data and the Future of AI | Cohere CEO Aidan Gomez

“But now across every single lab, the majority of the data is synthetic.”

Aidan Gomez · Synthetic Data and the Future of AI | Cohere CEO Aidan Gomez

“Like the models that they're building, potentially the best model, they may have surpassed OpenAI.”

Aidan Gomez · Synthetic Data and the Future of AI | Cohere CEO Aidan Gomez

Report last updated: Apr 21, 2026

Get started free

Back to Entities Intelligence Report