Tech 5: CoreWeave Plans US$4 Billion IPO, Trump Threatens CHIPS Act

1. CoreWeave plans IPO, faces Microsoft contract concerns

CoreWeave filed for a New York initial public offering (IPO) on Monday, seeking to raise US$4 billion and an expected valuation of more than US$35 billion.

On Wednesday, the Financial Times reported that Microsoft (NASDAQ:MSFT) pulled out of some of its agreements with CoreWeave. Anonymous sources didn’t give details as to why the startup’s biggest customer cancelled some contracts but alluded to Microsoft’s reduced confidence in CoreWeave after the company allegedly missed deadlines and ran into other delivery issues.

CoreWeave generates over 60 percent of its revenue from Microsoft, to which it supplies computing power from its data centers for running large-scale AI models, including OpenAI’s ChatGPT.

This multi-billion-dollar partnership represents a concentration risk. In its filing, CoreWeave stated that its business, operating results, financial condition and/or prospects could be negatively impacted by changes in its overall strategic relationship with Microsoft, including changes in demand and contractual agreements. Contracts between the two companies reportedly have Microsoft set to spend more than US$10 billion on CoreWeave services by 2030.

CoreWeave’s IPO filing revealed a US$1.9 billion revenue for 2024, alongside substantial debt and net losses. The company has raised US$14.5 billion through debt and equity financing, including US$11 billion in asset-backed loans. This aggressive expansion has led to escalating net losses, which reached US$863 million in 2024, up from US$594 million in 2023 and US$31 million in 2022.

The company’s reliance on chip supplier Nvidia (NASDAQ:NVDA) also poses supply chain risks, particularly concerning potential delays with Nvidia’s Blackwell GPUs.

After publication, CoreWeave delivered a statement to Data Center Dynamics, clarifying “there have been no contract cancellations or walking away from commitments. Any claim to the contrary is false and misleading.”

In a strategic move to further solidify its position in the AI space, on Tuesday, CoreWeave announced that it would acquire AI development startup Weights and Biases. The press release did not say how much the deal was worth, but unnamed sources for The Information said the deal could be valued at around US$1.7 billion.

2. TSMC fluctuates amid investment and political concerns

An interplay of factors, including geopolitical tensions and economic uncertainty, contributed to fluctuating TSMC’s (NYSE:TSM) share prices this week, both in the US and Taiwanese markets.

US shares were down at the start of the week due to concerns of economic upheaval and a potential trade war with China. Its Taiwanese shares fell after the company announced a US$100 billion investment in US chip production, including three new manufacturing plants, two packaging facilities and a research and development center.

Trump’s intention to end the US$52 billion CHIPS Act, which he expressed during his Tuesday evening Congressional Address, added to investor concerns. The CHIPS Act, an initiative from the Biden administration, has pledged funding to TSMC as well as fellow benefactors Intel (NASDAQ:INTC), Samsung (KS:5930) and Micron (NASDAQ:MU) to fund sizeable infrastructure projects. Intel received the largest portion, a US$7.9 billion grant to support commercial factories and another US$3 billion to produce military chips. TSMC is set to receive US$11.6 billion in direct funding and loans.

TSMC’s CEO, C.C. Wei, held a press conference on Thursday to address concerns from Taiwanese critics of the planned US investment who worry that moving advanced manufacturing will lessen US incentive to defend Taiwan from a Chinese invasion. The country’s Chinese Nationalist Party, the KMT, said the investment was a threat to national security.

Wei defended the move, stating it was a response to increased customer demand for AI chips. In a separate statement, Taiwan’s Economics Minister said that TSMC’s most advanced processes would stay in Taiwan until at least 2026.

He did not confirm whether Trump had guaranteed the continuation of CHIPS Act subsidies in light of the new investment pledge but said that the company could proceed without them, emphasizing the desire for fairness.

3. NVIDIA chips to power OpenAI and Oracle’s Stargate data center expansion

A source for Bloomberg said that OpenAI and Oracle (NYSE:ORCL) are preparing to add 64,000 of NVIDIA’s GB200 semiconductors to a new data center being built in Abilene, Texas, the first of the US$100 billion Stargate project announced by the Trump administration in January.

According to the report, the chips will be added to the center in phases, with an initial 16,000 chips set to be completed by this summer and the entire project complete by 2026.

4. Tech stocks share mixed earnings results

5. Shift to practical AI continues with agents, specialized applications

Key developments this week signaled a continuing shift toward AI agent expansion across both commercial and government sectors.

On Tuesday, Reuters reported on a new division from Amazon (NASDAQ:AMZN) Web Services (AWS) dedicated to AI agents, indicating a strategic focus on automated task solutions for cloud computing clients. The plans were officially announced by Amazon Vice President of AI and Data Swami Sivasubramanian via a LinkedIn post on Wednesday.

“This new capability – powered by Claude 3.7 Sonnet, Anthropic’s most intelligent model to date – allows developers to have more collaborative, interactive conversations with Q Developer that works with them, asks them feedback and makes iterative changes as they go along,” Sivasubramanian wrote.

Later, during a public interview at Morgan Stanley’s Technology, Media and Telecom Conference in San Francisco on Wednesday, Meta’s (NASDAQ:META) chief product officer Chris Cox said the company’s upcoming Llama 4 model will have reasoning capabilities powerful enough to create AI agents capable of using a web browser and other tools.

He described how more advanced AI agents can be built on a foundation of embeddings, enabling them to complete specific business-related tasks like filing receipts. These comments follow a previous CNBC report of Meta’s plans to debut a stand-alone AI app sometime during the second quarter and echo similar statements made to CNBC’s Julia Boorston by Clara Shih, Meta’s head of business AI.

“We’re quickly coming to a place where every business, from the very large to the very small, they’re going to have a business agent representing it and acting on its behalf, in its voice — the way that businesses today have websites and email addresses,” Shih said, explaining that Meta is working to develop business AIs for smaller businesses who may not be able to hire large AI teams.

Adding to this trend, OpenAI is reportedly planning to introduce tiered subscriptions for specialized AI agents, with prices ranging from US$2,000 to US$20,000 per month to reflect varying levels of capabilities.

Also, the US Department of Defense has begun integrating AI agents through collaborations with Scale AI, Microsoft, and Anduril for military operations, including simulation and decision support.

These moves signal rapid growth in the adoption of AI agents, marking a shift toward practical AI implementation and coincide with broader market shifts showing increased investment in AI applications, as noted in recent financial reporting from Bloomberg’s Kate Clark. This reflects a wider movement beyond foundational AI models, focused on delivering specialized, user-focused AI tools and services, whether through autonomous agents or dedicated applications.