Cerebras pulled off the biggest tech IPO in years, surging 68% on its first day, while both OpenAI and Anthropic launched consulting arms to embed AI engineers directly inside enterprises. The model race paused for breath; the deployment race sprinted.
The marquee event was Cerebras going public. The AI chipmaker priced its IPO at $185 per share — above the expected range — and opened at $350 on Nasdaq under the ticker CBRS on Wednesday, closing its first day up 68% at $311 and valuing the company near $95 billion. It raised $5.55 billion, making it the largest US tech IPO since Uber in 2019. The S-1 tells a revealing story about concentration: Cerebras reported $510 million in 2025 revenue, up from $290 million the year before, but a huge share of its forward pipeline rests on a single $20 billion-plus compute deal with OpenAI. AWS also plans to deploy Cerebras CS-3 systems in its data centres. The IPO arrived at a moment when every hyperscaler is trying to diversify its chip supply beyond Nvidia, and Cerebras is betting its wafer-scale architecture is the answer. Demand exceeded available shares by more than 20 times.
Both OpenAI and Anthropic made structural moves to own the enterprise deployment layer — and the two approaches were strikingly parallel. OpenAI launched the OpenAI Deployment Company, a majority-owned subsidiary backed by more than $4 billion from TPG, Advent, Bain Capital, Brookfield, Goldman Sachs, SoftBank, and others — 19 partners in total. It simultaneously acquired London-based AI consultancy Tomoro, bringing in roughly 150 forward-deployed engineers whose client list spans Fidelity, Virgin Atlantic, Tesco, and the NBA. The model: embed specialised engineers inside organisations to build and operate production AI systems. Anthropic countered by expanding its alliance with PwC, which will roll out Claude Code and Cowork to its global workforce, certify 30,000 US professionals on Claude, and launch a Claude-native finance business group. PwC reported concrete results: insurance underwriting cycles compressed from ten weeks to ten days, and cybersecurity incident response cut from hours to minutes. The consulting industry — long a middleman for technology adoption — is being refactored from both sides.
Anthropic also had a notably busy week on distribution. It launched Claude for Small Business, a package of 15 agentic workflows and connectors to QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, Microsoft 365, and Slack — targeting the tasks small businesses consistently dread: payroll planning, invoice chasing, month-end close, contract review. There's no extra charge beyond the Claude licence and whatever tools the business already pays for. Anthropic is also running a 10-city US tour offering free half-day AI training for 100 local business owners per stop — an unusually grassroots play for a frontier AI lab. Separately, the company announced a $200 million partnership with the Bill & Melinda Gates Foundation spanning global health, education, and economic mobility over four years. The health focus will target the 4.6 billion people who lack access to essential health services, while the economic strand will improve agricultural AI for smallholder farmers. It's a deliberate signal that Anthropic wants to be seen building public goods, not just enterprise value.
On the cybersecurity front, something genuinely new happened. Microsoft disclosed that its new AI scanning system, MDASH, found 16 of the 137 vulnerabilities fixed in this month's Patch Tuesday — four of them critical. The same week, Palo Alto Networks revealed it had used Claude Mythos and GPT-5.5 to deep-scan its entire product portfolio, discovering 75 issues across 130-plus products — compared to a normal monthly volume of fewer than five. Palo Alto published 26 CVEs in a single advisory, noting it was the first time the majority of findings came from AI rather than human researchers. None are being exploited in the wild, but Palo Alto warned that organisations have a three-to-five-month window before adversaries start using the same scanning techniques. The era of AI-discovered vulnerabilities is here, and patch cycles are about to get a lot heavier.
Regulation remained a rolling story. Fortune reported that US states have now introduced more than 1,200 AI-related bills, with at least 145 becoming law, creating a contradictory patchwork of requirements. Georgia signed an AI chatbot safety bill; Colorado sent four AI bills to the governor; California moved most of its AI legislation out of suspense. Meanwhile, the White House — prompted by the Mythos shock — is reportedly considering an executive order to establish an FDA-like pre-release vetting system for advanced AI models, a striking departure from the administration's earlier light-touch posture. The tension between federal coordination and state experimentation shows no sign of resolving, and for any company building AI products, compliance is becoming a multi-jurisdiction puzzle with no clean answer.