Cohere Labs heads to ICML Seoul as AI research becomes the sales floor

Why it matters

ICML is where AI founders convert research status into recruiting, customer access and infrastructure demand. The Seoul program will separate credible technical momentum from conference-week marketing.

Aidan Gomez, Nick Frosst and Ivan Zhang's Cohere Labs arrives in Seoul this week as ICML 2026 opens at the COEX Convention & Exhibition Center, putting frontier research, recruiting and AI infrastructure selling into the same six-day sprint.

The 43rd International Conference on Machine Learning runs July 6th through July 11th as an in-person conference in Seoul. ICML's own materials describe a main conference and workshops, with accepted papers either presented at the conference or included in the proceedings. Its downloads page lists 6,688 events or items, a scale that makes the conference less a single research meeting than a market map of what labs, founders and infrastructure vendors believe will matter next.

A post on X flagged Cohere Labs, Nebius and HPC-AI Tech among the company names around this year's program. The fully verifiable part is ICML itself, Cohere Labs' research presence as part of Cohere's broader research organization, and HPC-AI Tech's sponsor-side push in Seoul. The same item named Nebius, but official ICML materials reviewed for this story did not surface a specific Nebius-authored ICML 2026 paper. Nebius still matters to the week because AI cloud economics are now inseparable from what gets trained, benchmarked and commercialized after a paper lands.

Gomez's presence in that mix carries a particular history. He co-founded Cohere in 2019 after work at Google Brain, where he was part of the research orbit that produced the transformer architecture. University of Toronto coverage from 2022 described Cohere's founding goal as making natural language processing accessible to ordinary developers rather than only specialists with PhDs, and quoted Gomez explaining that he had seen how lack of resources and scarce talent kept most people from applying NLP systems. That origin story now sits inside a much tighter enterprise race: Cohere is selling models and workplace AI to regulated customers while its research arm works to keep the company visible in the places where technical credibility is minted.

The conference is where credibility gets converted

ICML is no longer only a venue for academic publication. A paper can help recruit researchers, validate a product thesis, support sales conversations with technical buyers and give a founder a reason to meet customers who would ignore a standard outbound pitch. ICML also remains a peer-review filter. A booth, workshop or company blog post around the conference should never be treated as proof of product superiority or customer traction. It is evidence of where a company wants to be seen.

For Cohere, that visibility comes after a large financing that is old news by conference standards but still frames the company's Seoul posture. Cohere said on August 14th, 2025 that it raised $500 million at a $6.8 billion valuation to accelerate enterprise and sovereign AI work. Cohere's own about page lists Gomez as co-founder and CEO, Frosst and Zhang as co-founders, Joelle Pineau as chief AI officer, Francois Chadwick as CFO, Frank O'Dowd as chief revenue and commercial officer, and Phil Blunsom as CTO. The same page lists offices including Toronto, New York, London, San Francisco, Montreal, Paris and Seoul, giving the company a local footprint in the conference city.

The tension for Gomez and Cohere is straightforward. Cohere's founding pitch was access: lower the bar so more developers could build with NLP. Its commercial focus has moved toward enterprise and regulated-sector deployment, where trust, control and procurement cycles matter as much as model benchmarks. ICML gives Cohere Labs a way to stay anchored in research while the company sells buyers on production AI. That is a useful balance, and a hard one to maintain as model companies compete with OpenAI, Anthropic, Google DeepMind and Mistral on one side, and with open-source distribution pressure on the other.

Cohere also enters the week under legal scrutiny. In February 2025, major publishers including Conde Nast sued Cohere, alleging copyright and trademark infringement tied to training and outputs, according to Ars Technica's report on the complaint. That lawsuit does not decide the technical quality of Cohere Labs' research. It does shape the backdrop for any enterprise AI company claiming trust as a core selling point.

HPC-AI Tech is selling the picks and shovels

HPC-AI Tech is taking the more explicit commercial route in Seoul. The Singapore-based AI infrastructure company says in a conference post that it will meet attendees at Booth B301, discuss LLM training, fine-tuning, AI agents, inference at scale, GPU infrastructure and cost-performance optimization, and run an incentive program for accepted ICML authors involving cloud GPU and model API credits.

That is a direct founder-market move: put compute credits in front of the authors most likely to need scale after publication. HPC-AI Tech's own company page says it develops Colossal-AI, an open-source project at the core of its platform, and that its software accelerates deep-learning training and inference on supercomputers and cloud platforms. The page lists Prof. Yang You as founder and says he received his PhD in computer science from UC Berkeley.

HPC-AI Tech's pitch also shows how research conferences have become distribution channels for AI infrastructure vendors. A paper author may need GPUs, fine-tuning tools, model APIs or a reserved cluster before they need a sales demo. By offering credits and technical conversations at ICML, HPC-AI Tech is trying to meet researchers at the moment their work begins turning into a system, company or lab program.

Nebius is the infrastructure shadow over the program

Nebius does not need a verified ICML paper to be relevant to the week. The Amsterdam-headquartered, Nasdaq-listed AI cloud company, led by founder and CEO Arkady Volozh, is part of the infrastructure class trying to absorb demand created by exactly the kind of research showcased at ICML.

Nebius describes itself as an AI cloud company spanning data, model training, tuning, runtime and deployment. Its about page says Volozh is founder, CEO and board member, and says Nebius is headquartered in Amsterdam and listed on Nasdaq. The company reported Q1 2026 revenue of $399 million, up 684% year over year, along with adjusted EBITDA of $129.5 million. Those are company-reported figures, and the same release notes that results include Nebius' core AI cloud business along with Avride and TripleTen.

Volozh's story is unusual in the AI infrastructure boom. Nebius emerged after the Dutch-listed Yandex holding sold its Russian assets in 2024 and rebuilt around AI infrastructure. That makes Nebius less a clean-room startup story than a public-company reinvention around the GPU supply squeeze. ICML is where that bet gets pressure-tested indirectly: if the research agenda keeps moving toward larger agent systems, multimodal models, scientific AI and test-time inference, cloud providers need to keep proving they can supply capacity without becoming a financing story detached from customer demand.

The competitive pressure is visible across the market. Together AI said on July 1st, 2026 that it raised an $800 million Series C at an $8.3 billion valuation, another sign that AI infrastructure remains one of the most heavily financed parts of the stack. For founders at ICML, that capital can look like cheaper credits, more model-hosting options and more vendors chasing their workloads. For the vendors, the hard part is converting conference visibility into durable usage once the credits run out.

The founder read on Seoul

The founders with the most to gain at ICML are not only the ones with accepted papers. They are the ones who can translate research trust into distribution without overstating what a paper proves.

For Gomez and Cohere Labs, Seoul is a chance to keep Cohere's original developer-access thesis connected to serious research while Cohere sells enterprise AI. For Yang You and HPC-AI Tech, ICML is a chance to turn author attention into infrastructure adoption. For Volozh and Nebius, the conference is part of the demand signal behind a public-market bet on AI cloud capacity.

The week will produce papers, awards, workshops and plenty of company claims. The useful read is narrower: which founders use ICML to recruit talent, earn technical credibility and attach their products to real research workflows. That is where a conference program starts becoming a company strategy.