Six developments in AI infrastructure constraints, private-market concentration, regulatory reset, and geopolitical competition.

Anthropic, the company that makes the Claude AI assistant, has agreed terms on a $30 billion fundraising round that would value it at roughly $900 billion before the new capital, according to the Financial Times, with the deal confirmed independently by The Information, the Wall Street Journal and Reuters. The round is co-led by four investment firms – Dragoneer, Greenoaks Capital, Sequoia Capital and Altimeter Capital – each committing $2 billion or more. As of the May 16–18 reporting window the deal had not been publicly announced by Anthropic, and Bloomberg's earlier May 12 reporting had stated no term sheet was yet signed; the FT's later reporting described terms as agreed.

The revenue figures attached to the round vary meaningfully by source and accounting method. Anthropic's last officially disclosed figure, from its February 2026 Series G, was a $14 billion annualized run-rate. The Wall Street Journal reported Anthropic is forecasting a $50 billion run-rate by mid-2026, up from ~$9 billion at the end of 2025; the FT reported annualized revenue is on track to top $45 billion "soon." Run-rate revenue annualizes current sales and is not the same as audited annual revenue, and Anthropic reports cloud-reseller revenue on a gross basis, which inflates the top line relative to net-reporting peers. Claude Code, Anthropic's coding tool, is separately reported to generate more than $2.5 billion in annualized run-rate revenue. Three of the four co-leads – Dragoneer, Sequoia and Altimeter – also hold significant stakes in OpenAI, which was valued at $852 billion in its March 2026 round. An IPO as early as October 2026 has been reported as under consideration.

Axios reported on May 18, citing Heatmap Pro data, that a record number of US data centers were canceled in the first quarter of 2026, as community and political resistance to AI infrastructure buildout intensifies ahead of the US midterm elections. Morgan Stanley analysts warned in coverage cited by Axios that public pushback could become a binding constraint on sector growth – not a passing communications problem. The cancellation wave sits against a wider supply picture: Sightline Climate data, cited by Bloomberg and summarized by TechRadar, puts the full announced US data center pipeline for 2026 at ~16 gigawatts, with only ~5GW actually under construction. That implies 30% to 50% of the 2026 pipeline is on track to be delayed or canceled.

Bloomberg reporting identified the binding constraints as transformers, switchgear, batteries and grid interconnection capacity – not funding and not semiconductors. A modern AI data center can be built in under 18 months, but the power infrastructure to feed it can take up to five years. The Federal Energy Regulatory Commission (FERC), the US government body overseeing electricity markets, is expected to act by the end of June on rules governing large data center grid connections, including how costs are assigned to developers. The International Data Center Authority found data centers now consume 2% of global electricity, up from 1.7% in 2024, and that political pushback typically intensifies once data centers exceed 5% of national electricity consumption – a threshold the US is near.

OpenAI, the company behind the ChatGPT assistant, launched a personal-finance experience in ChatGPT on May 15, available in preview to ChatGPT Pro subscribers in the US on web and iOS. Users can connect financial accounts through Plaid – a service that links to more than 12,000 institutions including Chase, Fidelity, Robinhood, American Express and Capital One – and see a dashboard of portfolio performance, spending, subscriptions and upcoming payments. Connected accounts are read-only: ChatGPT can see balances, transactions, investments and liabilities but cannot see full account numbers or move money. Intuit, the company behind TurboTax and QuickBooks, is the next planned integration. The launch was framed by American Banker as a direct challenge to banks' customer relationships.

The product runs on GPT-5.5 Thinking by default. OpenAI says it built an internal personal-finance benchmark with input from more than 50 finance professionals: GPT-5.5 Thinking scored 79 out of 100 and GPT-5.5 Pro scored 82.5 – a disclosure more rigorous than a typical model-launch claim, though not third-party validated. OpenAI reported more than 200 million users already ask financial questions in ChatGPT each month. The strategic core is a planned "answer-to-action" pipeline: through the Intuit partnership, a conversation about tax implications is intended to roll into a tax estimate and a scheduled session with a live local expert, and a credit-card recommendation into an approval-odds estimate and an application – all inside ChatGPT. The preview is US-only on the $100/month Pro tier; the $20/month Plus tier does not yet have access.

EU negotiators from the Council, Parliament and Commission reached a provisional agreement on May 7 on the Digital Omnibus on AI – the first set of amendments to the EU AI Act since its 2024 adoption. Detailed legal analysis continued through the May 16–18 window, including a breakdown published May 18 by Covington's Inside Privacy practice. The package combines timeline relief, targeted simplification and a set of substantive policy changes that are not purely deregulatory.

The headline change is a staggered deferral of compliance deadlines. Obligations for use-based high-risk AI systems – listed in Annex III of the AI Act, covering areas such as employment screening, credit scoring and biometric identification – are postponed from 2 August 2026 to 2 December 2027, a 16-month delay. Obligations for product-regulated high-risk systems in Annex I, covering items such as medical devices and machinery, move from 2 August 2027 to 2 August 2028. The deferrals reflect the practical difficulty of operationalizing rules before EU standards bodies finish the underlying technical standards. However, the co-legislators also added two new prohibited practices to Article 5: the use of AI to generate or manipulate non-consensual intimate imagery, and AI-generated child sexual abuse material. Both prohibitions take effect 2 December 2026. The grace period for providers to implement transparency solutions for AI-generated content was cut from six months to three, also landing on 2 December 2026. Formal adoption is expected in June, with publication in July.

DeepSeek, a Chinese AI lab known for releasing highly efficient open-weight language models at low cost, is in advanced talks to raise up to 50 billion yuan (approximately $7.35 billion) in its first-ever external funding round, according to reporting from Reuters, the Financial Times, The Information and the South China Morning Post. The round would value the previously self-funded lab at roughly $45–50 billion – a figure that escalated from initial discussions of ~$10 billion in under a month, a roughly fivefold jump. Some sources describe the current tranche as $3–4 billion against the higher ceiling; DeepSeek has not publicly confirmed the round. The lead investor is reported to be China's National Artificial Intelligence Industry Investment Fund, an $8.8 billion state-backed vehicle established in early 2025, with the China Integrated Circuit Industry Investment Fund (known as "Big Fund III") also involved.

Big Fund III was previously created to advance Chinese semiconductor self-sufficiency and has backed chipmakers SMIC and Yangtze Memory; its involvement with DeepSeek would be its first known investment in a Chinese large language model company. DeepSeek released its V4 model on April 24 – a 1.6-trillion-parameter system engineered to cut the memory and compute costs of long-context AI – and gave Huawei early access to optimize for V4 while reportedly declining that access to Nvidia and AMD. A V4.1 model is expected in June 2026. The stated trigger for opening the round is mundane: founder Liang Wenfeng opened it to grant employees equity after rivals began poaching DeepSeek researchers. The strategic subtext is that Beijing now appears to treat frontier AI software and domestic silicon as a single strategic problem.

CNBC reported on May 18, citing International Energy Agency data, that surging European energy prices risk undermining the continent's ambition to compete with the US and China in AI. The average price per megawatt of electricity in May was $111.65 in the UK, $88.97 in Germany, $44.19 in France and $28 in the US – meaning UK power costs roughly four times the US level and German costs roughly three times. The IEA separately found that energy-intensive industries in Europe paid on average about double US prices last year, and 50% more than in China and India.

The cost gap is already shaping where capacity gets built. CBRE research cited by CNBC found the cost of securing data center capacity in Europe's five largest markets – Frankfurt, London, Amsterdam, Paris and Dublin – is set to rise 12% in 2026. An Nvidia program manager told a Denmark conference that "the middle part of Europe has already lost the game," citing high electricity costs in Germany and the UK. Experts told CNBC that investment is likely to migrate within Europe toward lower-cost power regions, notably France with its nuclear base, creating winners and losers across the continent. The story connects to the same infrastructure constraint documented in Story 2: the International Data Center Authority report cited by CNBC found data centers now consume 2% of global electricity, with the US near the 6% national-consumption mark and political pushback typically intensifying above 5%.