The Trillion-Dollar Tug-of-War

The One Thing: Google and Amazon are now collectively committed to investing up to $73 billion in the same company, Anthropic, while competing with each other to be its primary distribution channel. This is not investment. This is a bidding war for the right to be the platform layer underneath intelligence itself.

If You Only Read One Thing: The Euronews deep dive on Anthropic's rapid ascent to an $800 billion secondary valuation explains why two cloud giants are racing to write checks that individually exceed the GDP of most countries.

TL;DR: Google committed up to $40 billion in Anthropic, just days after Amazon pledged up to $25 billion more, creating an unprecedented bidding war between cloud platforms for the same AI company. Beijing is retaliating against the Meta-Manus deal by ordering Chinese AI startups to reject American capital, severing the last integration channel between the two tech superpowers. And the DOJ dropped its criminal probe of Fed Chair Powell, clearing Kevin Warsh's confirmation path to replace him.

Google's $40 Billion Anthropic Bet: When Both Distributors Want to Own the Supplier

Google is investing up to $40 billion in Anthropic. $10 billion in cash immediately at a $350 billion valuation, with an additional $30 billion contingent on performance milestones. As Bloomberg first reported, this comes on top of Google's prior $3 billion in investments, bringing Google's total potential commitment to roughly $43 billion.

This landed four days after Amazon pledged an additional $5 billion, with up to $20 billion more conditional on milestones, atop its prior $8 billion. Amazon's total potential commitment: $33 billion. Combined, the two largest cloud providers in the world have committed up to $73 billion to the same company.

Meanwhile, on Forge Global's secondary market, Anthropic shares are now trading at an implied $1 trillion valuation, 2.6 times the $380 billion at which Anthropic closed its $30 billion Series G in February. That $1 trillion figure now exceeds OpenAI's $880 billion secondary price.

Why it matters (Platform Economics): The structural story here is not about Anthropic's models. It is about who controls the distribution layer for AI.

Google and Amazon are not making financial investments in the traditional sense. They are making strategic purchases of distribution rights. Amazon's deal locks Anthropic into $100 billion in AWS spending over a decade. Google's deal almost certainly includes comparable GCP commitments, though the terms are not public. Both companies are converting cash into guaranteed infrastructure revenue while securing privileged access to Claude models for their respective cloud customers.

This creates a dynamic that has no real precedent in technology. The two dominant cloud platforms are bidding against each other for the same supplier, each trying to ensure their cloud becomes the default home for Claude workloads. Anthropic is not choosing a side. It is playing them against each other, extracting capital from both while maintaining optionality.

The numbers reveal the tension. Anthropic's gross margins sit at roughly 40%, far below typical software margins. Its annualized revenue run rate surged from $9 billion at the end of 2025 to $30 billion by March 2026 (a 233% increase in one quarter), but CEO Dario Amodei has publicly warned that the company is expanding operations so fast it may not grow revenue quickly enough to keep pace with costs. Inference costs surged 23% above expectations in 2025, destroying 10 percentage points of gross margin.

Both Google and Amazon are, in effect, paying Anthropic to spend their own money back on their infrastructure. It is a closed loop: invest cash, receive guaranteed infrastructure revenue, subsidize a model provider that makes your cloud more attractive. The question is whether Anthropic captures the value of being the model layer, or whether it becomes a high-cost, low-margin utility subsidized by cloud giants who capture the real margin on compute.

Room for disagreement: Anthropic's revenue growth is genuinely extraordinary. A company that grew 10x annually for three consecutive years has earned the right to command premium valuations. And the dual-cloud strategy may be brilliant rather than dangerous. Being essential to both AWS and GCP gives Anthropic bargaining power that no other AI company possesses. If either cloud tries to squeeze Anthropic on terms, Anthropic can credibly threaten to shift workloads to the other.

What to watch: Whether Anthropic files for an IPO before year-end. The $1 trillion secondary valuation creates enormous pressure to give early investors and employees liquidity, and a public listing would finally establish a market-determined price for what may be the most valuable private company in history. Watch the gap between primary and secondary valuations. If it widens past 3x, it signals either irrational exuberance in secondary markets or that the next primary round will reprice dramatically upward.

Beijing Closes the Capital Door: The Last Channel of US-China Tech Integration Is Shutting Down

Chinese regulators have instructed the country's most prominent AI startups to reject American investment without explicit government approval. As Bloomberg reported, the National Development and Reform Commission told firms including Moonshot AI (which is considering an IPO) and StepFun to reject capital of US origin in upcoming funding rounds.

The trigger was Meta's $2 billion acquisition of Manus, the Chinese AI agent startup, which Beijing is now probing for illegal foreign investment and technology exports. The broader context: the US Treasury Department finalized restrictions last year limiting American investment in Chinese quantum computing, semiconductors, and AI. China's response is the mirror image.

Why it matters (Second-Order Effects): The US-China tech relationship has been decoupling in layers for years. Export controls restricted the flow of advanced chips. Entity lists restricted the flow of technology. Visa restrictions limited the flow of talent. But capital was the last open channel, the final mechanism through which American venture firms could fund Chinese AI startups and vice versa. That channel is now being closed from both sides simultaneously.

The symmetry matters. When only the US imposed outbound investment restrictions, Chinese startups could still raise from US-based firms willing to navigate compliance. Now China is imposing its own inbound restrictions, making the decision for them. For firms like Sequoia, which already split its US and China operations, there is nowhere left to bridge.

The Manus deal is the precipitating event, but Beijing's concern runs deeper. Chinese regulators are worried about a pattern: Chinese-founded startups exploring international opportunities, taking Western money, and potentially exporting homegrown technology in the process. The NDRC's intervention signals that "sovereign capital" is now a policy objective, not just a preference. Beijing is betting that restricting US money will slow some startup growth but produce a more aligned and controllable tech sector.

The second-order implication cuts against both countries. Chinese AI startups lose access to the deepest capital markets in the world. American investors lose access to what has been the second-most-productive AI ecosystem. DeepSeek's V4 Pro, which launched this week as a 1.6-trillion-parameter open-source model competitive with GPT-5.5, was built entirely with Chinese capital and Chinese-accessible chips. The restriction may accelerate, not slow, China's push for complete AI self-sufficiency.

Room for disagreement: Some analysts argue that complete economic decoupling is not actually occurring despite these restrictions. Trade volumes between the two countries remain massive. The NDRC restrictions may be selectively enforced, applying to high-profile AI companies while leaving less sensitive sectors untouched. And Chinese startups can still raise from non-US investors, including Middle Eastern sovereign wealth funds that have been aggressively funding Chinese AI.

What to watch: Whether the NDRC guidance becomes formalized regulation. Informal guidance can be reversed quietly. Formal law cannot. Watch Moonshot AI's IPO filing for whether it excludes US-domiciled investors, which would be the clearest signal that this is policy, not posture.

The Contrarian Take

Everyone says: Google's $40 billion investment proves Anthropic is the undisputed AI winner. At a $1 trillion secondary valuation, Anthropic has overtaken OpenAI as the most valuable AI company. The market has spoken.

Here's why that's incomplete: The $73 billion in combined hyperscaler commitments creates a dependency that may be Anthropic's biggest strategic vulnerability, not its strength. At 40% gross margins and a $19 billion annual burn rate, Anthropic cannot survive without continuous capital infusions from the very cloud providers it competes with for AI customers. Google Cloud sells its own Gemini models. AWS sells its own Bedrock-optimized offerings. Both are simultaneously Anthropic's largest investors, largest infrastructure suppliers, and most dangerous competitors. History suggests that when your distributor is also your competitor and your investor, the terms eventually shift in their favor. Ask any consumer brand that became dependent on Amazon's marketplace. The secondary market is pricing Anthropic as an independent platform company. The capital structure says it is a dependent one.

Under the Radar

The White House fired the new head of the Center for AI Standards and Innovation after four days because he previously worked at Anthropic. Collin Burns, a UC Berkeley PhD who studied AI safety at OpenAI and then Anthropic, was pushed out by the White House after officials learned of his background. Chris Fall, a DOE veteran from the first Trump term, replaced him. The Anthropic-government relationship has deteriorated to the point where merely having the company on your resume is disqualifying for federal AI roles.
Consumer sentiment hit a 74-year record low in the same week the S&P 500 and Nasdaq set all-time highs. The University of Michigan's final April reading landed at 49.8, the weakest in the survey's history, with year-ahead inflation expectations surging to 4.7%. This is the widest sentiment-market divergence on record. Either consumers are wrong about the economy, or the stock market is pricing an AI future that has not yet reached Main Street.
Cognition AI is raising at a $25 billion valuation for Devin, its autonomous AI coder. Bloomberg reports the company is seeking hundreds of millions, up from a $10.2 billion round last September. Devin's ARR was $73 million last June. At 342x revenue, the AI coding bubble is pricing growth that would require every developer on earth to subscribe.

Quick Takes

The DOJ dropped its criminal investigation of Fed Chair Jerome Powell, clearing the path for Kevin Warsh's confirmation. The probe, launched in January over $2.5 billion in Fed headquarters renovation cost overruns, was terminated by U.S. Attorney Jeanine Pirro. Sen. Thom Tillis had blocked Warsh's confirmation until the investigation ended. Powell's term expires May 16. The Fed inspector general will continue reviewing the renovation, preserving the option to reopen the case. (NPR)

X-energy surged 27% on its Nasdaq debut, raising over $1 billion in the largest nuclear IPO in history. The Amazon-backed small modular reactor company priced at $23 per share, well above its initial $16-$19 range, and closed at $29.20. This is the first advanced reactor company to pursue a traditional IPO rather than a SPAC. Competitor Oklo's stock fell on the debut, signaling that investors see X-energy as the category winner. AI data center energy demand is creating a new investable asset class in nuclear. (CNBC)

Nvidia closed at a record high, pushing its market cap past $5 trillion for the first time since October. Shares rose 4.3% to $208.27, making Nvidia worth $1 trillion more than the second-largest company, Alphabet. The rally was ignited by Intel's earnings, which proved that the AI chip spending cycle is broadening beyond Nvidia. But Nvidia's largest customers, Meta, Microsoft, and Amazon, are all developing custom AI chips that bypass Nvidia's order backlogs. (Yahoo Finance)

The Musk v. Altman trial begins Monday in Oakland, with witnesses including Ilya Sutskever and Satya Nadella. Musk's $134 billion lawsuit alleges Altman and Greg Brockman betrayed OpenAI's nonprofit founding mission. Musk seeks their removal and to unwind OpenAI's for-profit conversion. The trial lands as Musk prepares SpaceX's IPO and OpenAI gears up for its own public offering, making the outcome materially relevant to two of the largest expected listings of 2026. (Washington Post)

Stories We're Watching

The AI Capital Arms Race: Google vs. Amazon for Anthropic (Week 1) — Google's $40 billion and Amazon's $33 billion total commitments to the same company create an unprecedented bidding war for AI distribution rights. Watch for whether Microsoft responds with a comparable commitment to OpenAI or pivots to building in-house.
US-China Capital Decoupling: Both Doors Closing (Week 1) — Beijing's NDRC ordering startups to reject US money, combined with US outbound investment restrictions, is severing the last integration channel. If Moonshot AI's IPO excludes US investors, the split becomes structural.
Musk v. Altman: The AI Trial of the Century (Day 0) — Jury selection Monday. A Musk victory could force OpenAI to unwind its for-profit conversion, fundamentally reshaping the AI industry's corporate structure. Even a loss surfaces internal documents that could damage OpenAI's IPO narrative.

The Thread

The thread connecting today's stories is the question of who captures the value in AI. Google and Amazon are spending $73 billion to ensure they control distribution. China's NDRC is ensuring its government controls which capital flows into its AI ecosystem. The DOJ dropped a politically motivated investigation to install a Fed chair more sympathetic to the administration's economic agenda. X-energy is pricing the energy infrastructure that makes all of it possible. And on Monday, a federal court will decide whether OpenAI's original nonprofit structure was a binding promise or a fundraising strategy.

At every layer, from chips (Nvidia at $5 trillion) to energy (X-energy's $1 billion IPO) to models (Anthropic at $1 trillion secondary) to distribution (the hyperscaler bidding war) to regulation (Burns fired for his Anthropic affiliation), the question is the same: who gets to sit between AI capability and the people who use it? The trillion-dollar bets being placed this week are attempts to answer that question before the market does.

Predictions

New predictions:

I predict: Anthropic files for an IPO or announces IPO intention before the end of 2026. The 2.6x gap between primary ($380B) and secondary ($1T) valuations creates unsustainable pressure for employee and investor liquidity. The dual-cloud strategy gives Anthropic enough independence to go public without being captured by either backer. (Confidence: medium-high; Check by: 2026-12-31)
I predict: China's NDRC investment restrictions will be formalized into binding regulation within 90 days, applying to AI, quantum computing, and advanced semiconductor companies. The Manus probe provides political cover, and the restrictions are already being enforced informally. (Confidence: medium; Check by: 2026-07-25)

Coming Next Week

Next week opens with the biggest AI trial in history: Musk v. Altman begins Monday in Oakland, with Ilya Sutskever, Greg Brockman, and Satya Nadella all expected to testify. We will be covering the proceedings daily for what they reveal about OpenAI's founding intent, its for-profit conversion, and the structural question of whether AI companies owe a fiduciary duty to humanity or to shareholders. Also on the radar: Warsh's confirmation vote could be scheduled now that the Powell probe is cleared, and we are watching for Microsoft's response to Google's Anthropic move.

Generated: April 25, 2026, 5:45 AM ET