AI Intelligence Briefing – May 28, 2026 Issue #173

The House Finance Committee summons the OpenAI CEO as GOP lawmakers raise fiduciary and governance concerns ahead of the company's planned public listing.

May 28, 2026 | Reading time: 8 minutes | Issue #173

Sam Altman appeared before the House Financial Services Committee on Tuesday, fielding questions about OpenAI's governance structure, capitalization table, and the adequacy of disclosures related to its reported $300 billion valuation. The hearing was convened after a group of Republican senators sent a letter to the SEC questioning whether OpenAI's dual-class stock structure — which gives Altman supervoting control — meets the exchange's listing standards for governance accountability.

"When one individual holds 51% of voting power in a company valued at $300 billion, we need to understand who that individual answers to," said Representative Patrick McHenry, opening the hearing. The session was unusually bipartisan; progressive Democrats joined Republicans in expressing concern that OpenAI's governance model could set a precedent for other AI companies seeking public capital.

Altman's prepared testimony emphasized OpenAI's transition to a public benefit corporation and its commitment to AI safety benchmarks. He also disclosed new details about the company's capitalization: Microsoft holds a 49% stake, a group of venture investors holds 23%, and employees hold the remainder through options and RSUs. Altman himself holds no direct equity — a point he reiterated under questioning — but controls the voting stock through a holding entity.

The market reaction to the hearing was muted. OpenAI has not yet publicly filed its S-1, and the scrutiny is largely procedural. But the hearing signals that AI companies seeking public capital in 2026 will face a level of governance scrutiny that pure-play tech IPOs have not experienced in decades. The question is no longer whether AI companies can go public. It is whether their governance structures can survive the process.

Source: Reuters

Anthropic speeds IPO timeline to beat OpenAI to market

Anthropic announced on Wednesday that it has selected Goldman Sachs and Morgan Stanley as lead underwriters for a public offering targeted for October. The announcement comes less than 48 hours after OpenAI's congressional hearing, suggesting Anthropic's leadership sees an opportunity to reach public markets before their larger rival gets bogged down in Washington regulatory review.

Dario Amodei, Anthropic's CEO, told CNBC that the company "is not looking to replicate OpenAI's governance mistakes." Anthropic's board includes independent directors and a 10x voting cap on founder shares — a direct contrast to OpenAI's supervoting structure. The company's last private valuation was $61.5 billion, and early indications suggest an IPO could price between $75–85 billion.

The timing is aggressive. Anthropic has yet to report a full year of audited financials, and its revenue — while growing rapidly — is still heavily dependent on a handful of enterprise customers. Anthropic is essentially betting that reaching public markets before OpenAI is more important than perfecting the balance sheet. If the offering succeeds, it becomes the first major AI lab with public equity. If it struggles, it could chill the broader AI IPO pipeline.

Source: CNBC

HBM cost crisis spreads: AMD reports memory eating 61% of chip bill

AMD's Q1 2026 earnings, reported Wednesday, confirmed what Epoch AI flagged for NVIDIA: high-bandwidth memory is eating the chip. AMD's data center unit reported $4.8 billion in revenue, up 34% year-over-year, but gross margins contracted 180 basis points to 47%. CFO Jean Hu attributed the compression directly to "HBM pricing dynamics," noting that memory costs rose 22% quarter-over-quarter while logic die costs were flat.

"We are not NVIDIA," Hu told analysts. "We do not have the pricing power to pass component inflation through to customers who have alternatives."

The comment is significant. NVIDIA's 74.9% gross margins and hyperscaler lock-in allow it to absorb HBM inflation. AMD, Intel, and the custom silicon efforts at Google and Amazon do not have the same cushion. AMD's data center revenue is growing fastest of any division, yet it is the least profitable — a structural warning that the AI chip business is bifurcating into a NVIDIA tier and everyone else.

Source: AMD Earnings Call

China's DeepSeek-V4 draws 10M users in first 24 hours

DeepSeek reported that its V4 preview — released Monday — attracted over 10 million users in its first day of availability. The figure is notable for a model positioned as a reasoning-first system, not a general-purpose chatbot. DeepSeek's strategy appears to be positioning V4 as the "default" model for developers building agents and workflows, while competitors chase consumer engagement.

The release also drew attention to an architectural choice: DeepSeek-V4 uses a mixture-of-experts (MoE) architecture with 1.2 trillion total parameters but only 32 billion active per forward pass. The design keeps inference costs extremely low. DeepSeek is reportedly charging $0.08 per million tokens for API access — roughly one-third of GPT-4.5's pricing and one-fifth of Gemini 3.5's. The company told The Verge that it is "selling compute at cost" to gain market share.

The risk, as always, is sustainability. DeepSeek's burn rate on inference is likely massive at $0.08 pricing. But the user numbers validate a market thesis: developers are price-sensitive and will switch models for a 5x cost advantage, even if the capability gap is modest.

Source: The Verge

Google expands AI Mode to EU, faces regulatory inquiry

Google announced Wednesday that AI Mode — the agentic search interface unveiled at I/O — is now available in the European Union, the United Kingdom, and Japan. The expansion comes three weeks after the U.S. rollout, which Google claims changed "millions" of daily query patterns by surfacing synthesized answers instead of traditional blue links.

The EU launch triggered an immediate regulatory response. The European Commission's Directorate-General for Competition confirmed that it has opened a "preliminary investigation" into whether AI Mode constitutes an abuse of Google's dominant search position by self-preferencing its own AI-generated content over third-party publishers. French publishers' association Geste filed a formal complaint arguing that AI Mode's summarization effectively replaces visits to publisher websites, starving them of advertising revenue.

Google's defense, outlined in a blog post, is that AI Mode cites sources and includes links. Publisher associations counter that click-through rates from AI summary citations are "materially lower" than standard search results. The outcome of the EU inquiry will likely set the regulatory framework for how AI interfaces can ingest and republish third-party content across all platforms.

Source: Google Blog

Capital Flows

The capital story this week is a race between governance and growth. OpenAI and Anthropic are both sprinting toward IPOs, but for opposite reasons: OpenAI needs the capital to fund its multi-hundred-billion-dollar infrastructure buildout, while Anthropic needs the credibility of public-market governance. AMD's earnings confirm that the HBM cost squeeze is not a NVIDIA-specific problem — it is an industry-wide margin contraction that will disproportionately hit second-tier chip designers.

In China, DeepSeek's aggressive pricing strategy suggests a new model for AI distribution: sell at cost, capture developers, monetize later. It is the same playbook AWS used with EC2 in 2006. Whether DeepSeek has the balance sheet to sustain it is the open question.

The Miss

A Bloomberg story published Tuesday reported that several Fortune 500 companies are quietly building "shadow AI" budgets — unaudited lines of credit allocated to AI experiments that do not appear on official IT budgets. The practice is spreading because CIOs cannot get board approval for AI spending without proof of ROI, but cannot get proof of ROI without running experiments. The result is a growing black market in corporate AI spending that is invisible to investors, auditors, and regulators. The article was buried under earnings coverage. It deserves more attention.

Source: Bloomberg

Pull Quotes

"When one individual holds 51% of voting power in a company valued at $300 billion, we need to understand who that individual answers to." — Representative Patrick McHenry

"We are not NVIDIA. We do not have the pricing power to pass component inflation through." — Jean Hu, AMD CFO

"Selling compute at cost" to gain market share. — DeepSeek on V4 pricing strategy

Reads & Links

• OpenAI Congressional Testimony Transcript — Full text of Sam Altman's House FSC appearance. Reuters
• AMD Q1 2026 Earnings — Revenue up, margins down, HBM costs bite. AMD IR
• Anthropic IPO Timeline — Goldman and Morgan Stanley engaged; October target. CNBC
• DeepSeek-V4 User Numbers — 10M users in 24 hours; $0.08 per million tokens. The Verge
• Google AI Mode EU Investigation — European Commission opens preliminary probe. Google Blog
• "Shadow AI" Budgets — Fortune 500s hide AI spend from boards and auditors. Bloomberg