May 24 Roundup: OpenAI Files for IPO, Karpathy Joins Anthropic, and AI Cracks an 80-Year Math Problem
A week of seismic moves in AI: OpenAI sets its sights on a September stock market debut, Anthropic snags one of the field's most celebrated researchers and reports its first profitable quarter, Google's Gemini Omni model family rewires multimodal creation, Trump's White House shelves voluntary pre-release oversight, and an OpenAI reasoning model quietly dismantles a geometry conjecture that had stumped mathematicians since 1946. Here's what it all means for enterprise strategy.
1. OpenAI Confidentially Files for IPO, Targets September Debut
OpenAI is preparing to become a publicly traded company — and it's moving fast. According to reporting from Reuters and Bloomberg, the company is working with Goldman Sachs and Morgan Stanley on a draft IPO prospectus and plans to confidentially file with the SEC imminently, with a public market debut targeted for as early as September 2026.
The move represents a historic milestone for the company that helped ignite the modern AI era. OpenAI completed its conversion from a capped-profit nonprofit structure earlier this year, clearing the structural pathway to a public offering. The IPO would follow a $122 billion funding round that valued the company at roughly $852 billion — making it one of the most valuable private companies in history before even hitting public markets.
"OpenAI is aiming to go public as early as September and is working with Goldman Sachs and Morgan Stanley on a draft IPO prospectus that it plans to file with the regulators soon." — Reuters, May 20, 2026
The timing is notable: OpenAI's move comes in the same week Anthropic announced it is on pace for its first profitable quarter. Both companies are rapidly shifting from research organizations burning capital to mature commercial enterprises capable of sustaining themselves — and attracting public-market investors.
OpenAI has also recently been named a leader in enterprise coding agents in Gartner's 2026 Magic Quadrant, and its partnership with Dell Technologies to bring Codex to hybrid and on-premises environments signals the company's aggressive push into Fortune 500 infrastructure.
An OpenAI IPO in the fall changes the competitive dynamics significantly. Public markets will demand revenue clarity, margin improvement, and durable enterprise contracts — pressures that could slow product experimentation but accelerate enterprise-grade reliability and support. For enterprises building on OpenAI infrastructure, a public company is a more accountable partner than a heavily funded startup. Expect more formalized SLAs, enterprise licensing tiers, and a harder push into regulated industries where a public company's compliance posture matters. The September target also sets up a fascinating Q4: OpenAI will be running an IPO roadshow at the same time it's competing fiercely for enterprise deals with Anthropic, Google, and Microsoft.
2. Anthropic Scores Andrej Karpathy and Reports First Profitable Quarter
Anthropic had a banner week on two fronts. First, the company announced that Andrej Karpathy — co-founder of OpenAI, former director of AI at Tesla, and one of the most respected researchers in the field — is joining as a research leader. Then, the company told investors it will more than double revenue to approximately $10.9 billion in Q2 2026 and deliver an operating profit for the first time.
Karpathy's hire is a talent coup that carries deep symbolism. He helped build OpenAI from the ground up before Elon Musk poached him for Tesla's Autopilot program in 2017. His return to the frontier-model space — and specifically to Anthropic rather than back to OpenAI — signals where serious researchers believe the most interesting and responsible work is happening. He will build a team focused on using Claude to accelerate pretraining research.
"I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D." — Andrej Karpathy, post on X, May 19, 2026
On the financial front, TechCrunch and the Wall Street Journal report that Anthropic shared financials with investors showing more than doubled revenue heading into Q2. The milestone is significant: Anthropic was widely viewed as burning through capital with no near-term path to profitability. The shift reflects surging enterprise adoption of Claude, particularly among legal, financial, and technical professionals who have increasingly expressed preferences for Claude over competing models.
There's a catch: Anthropic may not sustain profitability throughout the year due to large scheduled compute costs — the company has committed to massive infrastructure buildouts including compute capacity at xAI's Colossus data center in Memphis. Still, hitting operating profit even once rewrites the narrative around Anthropic's viability.
Two things are happening simultaneously at Anthropic that should shape your vendor strategy: the company is becoming commercially viable AND attracting world-class research talent. That's a rare combination. Karpathy's focus on pretraining research suggests Anthropic is investing heavily in the next generation of model capabilities — not just polishing the current Claude family. For enterprises considering multi-year Claude commitments, this is a positive signal on both reliability and capability trajectory. The profitability milestone also reduces the "key-person" and "funding cliff" risks that make some enterprise buyers nervous about startups. Anthropic is beginning to look less like a bet on a startup and more like a bet on an institution.
3. Google's Gemini Omni Redefines Multimodal AI Creation
Google I/O 2026 gave the world its clearest view yet of where multimodal AI is heading — and the answer is "everywhere, from any input." The centerpiece of Google's launch week was the introduction of Gemini Omni, an entirely new family of AI models that can generate video clips from any combination of text, photos, existing video, and audio inputs simultaneously.
The first model in the family, Omni Flash, is already live in the Gemini app, Google Flow, and YouTube Shorts. Unlike Google's earlier Veo model — which was limited to text-to-video — Omni Flash accepts truly mixed-media inputs and is designed to eventually "create anything from any input," according to Google.
Google also launched Gemini 3.5 Flash as the new default model powering both the Gemini app and AI Mode in Search. The update brings meaningful improvements: faster response times, better agentic task handling, improved coding capabilities, richer UI generation, and enhanced content guardrails. Gemini 3.5 Pro is scheduled to follow next month.
Perhaps the most strategically significant announcement was Gemini Spark — Google's ambient AI agent that runs 24/7 in the background via virtual machines on Google Cloud, connected to Gmail, Docs, Sheets, and other Workspace apps. Spark can write emails, create study guides, monitor for hidden credit card fees, and proactively alert users when it finds relevant information. It represents Google's clearest move yet into the always-on personal AI layer that competitors like OpenClaw and Microsoft Copilot are also targeting.
"Google launched updated AI models at I/O, starting with Gemini 3.5 Flash... Google says the new model is significantly faster, better at handling agentic tasks, offers improved agentic coding capabilities, and generates 'richer, more interactive web UIs and graphics.'" — The Verge, May 19, 2026
Gemini Omni is a significant capability leap because it dissolves the modality boundaries that have made AI workflows fragmented. Enterprises currently running separate tools for video editing, image generation, and audio work will increasingly find a single Gemini-powered pipeline can handle all three from a single prompt chain. The business implication: content teams, marketing departments, and product design groups should be piloting Gemini Flow and Omni now — not waiting for Q4 budget cycles. Gemini Spark's ambient agent model is also a direct signal that Google is competing for AI infrastructure "ownership" within Google Workspace organizations. If your team lives in Gmail and Docs, Spark will increasingly look like the obvious always-on intelligence layer.
4. OpenAI's Model Disproves an 80-Year-Old Geometry Conjecture
In a development that startled the mathematics community, OpenAI announced that one of its internal reasoning models has produced an original mathematical proof disproving a famous unsolved conjecture in discrete geometry — one first posed by legendary mathematician Paul Erdős in 1946.
The problem, known as the "unit distance problem," concerns how many pairs of points in a set of n points can be exactly one unit apart. For nearly 80 years, mathematicians believed the optimal configurations looked roughly like square grids. OpenAI's model discovered an entirely new family of constructions that yields a polynomial improvement over grid-based arrangements — disproving the conjecture with a previously unknown approach.
What makes this result credible — unlike a premature claim last October about GPT-5 and Erdős problems that turned out to be hallucinated — is that OpenAI published companion remarks validating the proof from established mathematicians including Noga Alon, Melanie Wood, and Thomas Bloom, who maintains the Erdős Problems website.
"For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better." — OpenAI, post on X, May 20, 2026
Mathematicians who reviewed the result called it a genuine advance, with some noting that "AI has gone beyond being just an assistant" in describing the surprising nature of the discovery. The result suggests that AI reasoning models are beginning to produce genuinely novel intellectual contributions — not just pattern-matched solutions from training data.
This result matters less for its specific mathematical content and more for what it implies about AI's role in knowledge work. For years, AI has been framed as a retrieval and synthesis engine — useful for organizing what humans already know, but not capable of original discovery. The Erdős disproof challenges that framing directly. Legal professionals should watch the implications for prior art and patent research. Pharma R&D teams should note what it might mean for protein structure and drug discovery. Strategic planners should ask: if AI can find constructions in pure mathematics that humans missed for 80 years, what might it find when turned toward your product design, supply chain optimization, or competitive analysis? The answer may be uncomfortable — but it's the right question.
5. Trump Shelves Voluntary Pre-Release AI Oversight Plan
A proposed White House policy that would have created a voluntary oversight framework for advanced AI models has been blocked by President Trump, according to Politico. The policy would have allowed developers of frontier AI models to voluntarily submit their products to a review by federal agencies up to 90 days before public release.
Politico reports that Trump had "many concerns" about the draft order and that it was scrapped following internal deliberations. The administration's position appears to be that voluntary pre-release review — even with no enforcement mechanism — represents too much government involvement in private technology development.
The decision is notable because even the AI industry's leading companies had signaled support for some form of pre-release engagement with government agencies. Microsoft, Google, and xAI had previously agreed to provide the U.S. government with pre-release access to frontier models under a separate voluntary commitment. The blocked policy would have formalized and expanded that informal arrangement.
Meanwhile, Colorado's Governor signed Senate Bill 26-189 substantially revising the state's pioneering AI Act, and Connecticut's AI bill passed the legislature — continuing a trend of state-level AI governance advancing in the absence of a comprehensive federal framework.
The federal retreat from even voluntary pre-release oversight creates a vacuum that state governments are actively filling. Colorado, Connecticut, California, and Maryland have all taken meaningful legislative steps this spring. For enterprises operating nationally, this means compliance obligations will increasingly vary by state, with no federal floor to standardize against. Legal and risk teams need to be building a state-by-state AI compliance map now — not after a patchwork of laws takes effect. On the flip side, the federal light-touch stance means AI product development cycles face fewer obstacles at the national level, which could accelerate enterprise AI deployment timelines for organizations confident in their own internal governance.
6. AI Reshapes the Labor Market: Blue-Collar Workers Positioned to Win
As Cisco announced 4,000 job cuts explicitly attributed to AI automation and banks issued unusually candid reports about AI displacing workers, new data from CNBC and ADP payroll analysis offered a counter-narrative: workers in roles where AI augments their work — rather than automates it — are seeing employment growth, while skilled trade workers are emerging as potential winners in the AI economy.
CNBC's analysis of ADP payroll data found that the augmentation/automation divide is the key variable. Workers who use AI as a collaborator — think data analysts, design engineers, healthcare practitioners, skilled tradespeople — are seeing stronger employment and wage outcomes than workers in roles targeted for full automation. PwC's 2025 Global AI Jobs Barometer found that workers with demonstrated AI skills command wage premiums up to 56% higher than their peers in comparable roles.
Separately, a massive new study comparing over 100,000 people with today's most advanced AI systems found that generative AI can now outperform the average human on certain creativity tests — a benchmark that adds complexity to the "AI won't replace creative workers" narrative that has provided comfort to some knowledge workers.
"Those who held jobs where AI was poised to augment their work versus automate saw growing employment in the same time period." — CNBC, citing ADP payroll data analysis, May 19, 2026
The labor market signal is becoming clearer: the dividing line is not "uses AI" versus "doesn't use AI" — it's whether your role involves judgment, physical presence, or relationship context that AI cannot replicate without you. Enterprises have a short window to help their workforces move toward the augmentation side of that line before the automation pressure arrives. That means investing in AI literacy programs now, redesigning workflows so that human workers are paired with AI tools in ways that create more value than either alone, and building internal talent development programs for the people who will manage AI-augmented processes. The 56% wage premium for AI-skilled workers is also a recruiting and retention signal: enterprises that create those skills internally will have a structural advantage over those that must pay market rates to hire them.
7. EU AI Act Timeline Extended — But Stricter Rules Still Coming
The European Union has provided compliance relief for enterprises operating under the AI Act, postponing obligations for high-risk AI systems used in certain applications from August 2026 to December 2027 — a 16-month extension. The EU's approach reflects a two-tiered strategy: providing breathing room on use-based high-risk AI systems while maintaining or tightening timelines for general-purpose AI models and outright prohibited applications.
The extension is particularly relevant for enterprises deploying AI in HR, finance, healthcare, and other regulated sectors classified as high-risk under the Act's Annex III. Organizations that had been scrambling to meet the original August 2026 deadline now have additional time to build out conformity assessments, transparency documentation, and human oversight systems.
However, compliance teams should note that the extension is targeted, not universal. General-purpose AI model obligations — covering the underlying models powering many enterprise applications — remain on the original schedule, and the Act's prohibitions on unacceptable-risk AI applications (like real-time biometric surveillance in public spaces) are not affected by the extension.
The EU extension is welcome relief, but it would be a mistake to treat it as a signal to slow down compliance work. The 16-month window is an opportunity to do compliance properly rather than reactively — mapping AI systems to risk tiers, building documentation pipelines, and establishing governance processes that will scale. Enterprises with significant EU operations should use the extension to move from "checking boxes before the deadline" mode to "building durable AI governance infrastructure." The organizations that get ahead of this now will have a competitive advantage when enforcement ramps up: they'll be able to deploy new AI capabilities faster because their governance processes are already mature. Use the time wisely.
Why This Week Matters for Your Business
This week's news isn't a collection of isolated headlines — it's a coherent picture of the AI industry entering its next phase. The commercial viability story is locking in: OpenAI is heading to public markets, Anthropic is profitable, and Google is shipping product rather than previewing it. The talent signal is clear: the best researchers are choosing where to be, and they're choosing based on who is doing the most meaningful work. The policy signal is mixed but critical: federal oversight is loosening while state and international frameworks are tightening. And the capability signal — an AI model disproving an 80-year-old math conjecture — is a reminder that the ceiling of what these systems can do keeps rising in ways that will continue to surprise even the most informed observers. Enterprises that are still in "pilot mode" need to make a decision: this is no longer a trial phase. It's infrastructure.
Need help navigating AI for your business?
From IPO-era OpenAI strategy to EU AI Act compliance and workforce transformation — our team turns these developments into actionable plans for your organization.
Contact SEN-X →