Good morning. We’re looking at how OpenAI new structure, how Anthropic is winning the enterprise crowd, and more (energy, regulation, etc.).
OpenAI turns into a $130 billion for-profit
After NVIDIA reaching $5T of valuation, the news of the week is: OpenAI has completed a sweeping restructuring into a for-profit Public Benefit Corporation valued at about $500 billion. The shift allows it to raise capital freely — and possibly pursue an IPO as early as 2026 — while keeping a non-profit parent (owning 26% of the for-profit) to oversee its mission.
This change allows OpenAI to remove the cap on financial returns, appealing to investors and paving the way for significant capital raising. The governance structure remains complex, with the nonprofit board overseeing the for-profit entity. Concerns from regulators and former employees highlight the ongoing tension between profit motives and ethical AI development.
At the same time, OpenAI quietly renegotiated its contract with Microsoft, its largest investor and cloud partner. The deal extends their partnership to 2032 and relaxes exclusivity terms. Microsoft keeps early access to OpenAI’s research models and, crucially, will now retain access even after OpenAI reaches artificial general intelligence (AGI) — provided safety guardrails are in place and verified by an independent panel.
A funny note: with 27% of the new for-profit PBC structure, Microsoft transformed it’s $13.8B investment in OpenAI into approximately $135 billion. Quite a good ROI.
OpenAI also gains freedom to work with other partners and has committed to buy $250 billion in computing power from Microsoft (some more ROI), while remaining free to use other clouds. Microsoft, meanwhile, loses rights to OpenAI’s new hardware prototypes — a line developed after acquiring the Jony Ive (former Apple master of design) design firm.
Why it matters: This corporate makeover blurs the line between public good and private profit. The new model could fund faster model development and global rollout, but it also raises the risk of over-reliance on one vendor for enterprise AI.
Enterprise AI meets a reality check, and Financial services lead the comeback
After a year of hype, the corporate AI rollout has entered what the Financial Times calls a “messy middle”. Roughly 95 percent of generative-AI pilots have failed to scale. Companies are discovering that the technology is easy to test but hard to embed - workflows, governance and data quality still block results.
More analysis also point that productivity claims may even be overstated: developers using AI tools often work slower, not faster. Yet beneath the noise, a clearer pattern is emerging, and it’s coming from financial services.
Claude steps into the boardroom
While OpenAI dominates headlines, Anthropic has quietly positioned its model Claude as the go-to system for banks and insurers. Its new Claude for Financial Services platform offers Excel plug-ins, market-data connectors, and pre-built “Agent Skills” for diligence, risk modelling and document review; and is thus gaining attention from City, Bridgewater, Commonwealth bank but also insurance.
At AIG, the system is already changing underwriting workflows. CEO Peter Zaffino told investors earlier this year: that Claude helped AIG compress review times by fivefold while boosting data accuracy from 75 to over 90 percent (although as we just discussed there is a need to be careful about such productivity claims).
Anthropic’s strategy is clear: become the B2B alternative; slower to market than ChatGPT, but easier to integrate for regulated industries. OpenAI is chasing scale; Anthropic is chasing fit.
Why it matters for Zurich:
The contrast between OpenAI and Anthropic shows the two paths of AI adoption — one driven by platform scale, the other by enterprise integration.
Zurich’s success will depend less on “using AI” and more on embedding it: into workflows, data pipelines, and regulatory frameworks. Something our partnership with HSG and ETH might help with.
As the failure rate of generic pilots rises, choosing vendors with domain-specific models and audit-ready workflows could sharply reduce risk. And perhaps Anthropic is the right player for that.
Sidenote: Zurichat is slowing working on integrating Claude into its service to reduce reliance on Microsoft (main ChatGPT provider).
More top news
Energy & Infrastructure
Google committed $1.6 billion to reopen a shuttered nuclear plant in Iowa to power AI data centers - the clearest sign yet that AI’s energy appetite is reshaping electricity markets.
Rising energy prices put AI in the crosshairs: Data centers already use 4 percent of U.S. electricity, and public backlash is mounting.
Regulators are increasingly focused on how data centers impact local communities and energy resources, with concerns about their strain on outdated power grids.
In the spam of two weeks, AWS then Microsoft Azure both suffered outage showcasing the massive risk of cloud failure and over reliance on a few giants.
Regulation & Policy
LinkedIn begins scraping EU user data for Microsoft’s AI training. Users in the EU, EEA, Switzerland, Canada and Hong Kong must opt out by today (Nov. 3).
Capgemini’s CEO calls for a pause on the EU AI Act, warning that the legislation risks locking Europe out of AI competitiveness.\
AI-generated fake receipts are on the rise and companies seem to be struggling to deal with it. A trend that could soon be seen in Insurance Claim too.
Start-ups & Funding
Mercor reaches a $10 billion valuation, supplying contractor expertise for AI model training.
Fireworks AI raises $254 million at a $4 billion valuation, focusing on inference efficiency — the “engine room” of large models.
The famous text-optimiser Grammarly is changing its name to SuperHuman. Another startup trying to diversify and not get eaten by OpenAI or Anthropic.
Tips & Tricks
If you are interest in the worst risk presented by the rise of AI, the AI 2027 scenario is the THE reference to read at least once. For the non-reader and less interested in the topic, I can only recommend this great video on the scenario from Aric Floyd: https://www.youtube.com/watch?v=5KVDDfAkRgc. Nothing about Terminator here, but a great and smart exploration of the geopolitical and economical impact of such technology and how it might reshape our world for the better or the worst.