Shares of CoreWeave rose more than 13%, extending an already strong run for the stock as investors welcomed another major customer win just a day after the company unveiled a $21 billion expansion of its cloud partnership with Meta Platforms.

The agreement with Anthropic, whose financial terms were not disclosed, will bring fresh compute capacity online later this year to support production-scale workloads for Claude, one of the fastest-growing frontier AI model families in the market. According to the companies, the partnership will begin with a phased infrastructure rollout, with scope for future expansion as demand accelerates.

The deal is believed to be more than another cloud-services contract. It is seen as a strategic supply agreement in an industry where access to compute has become as critical as access to capital.

The race among leading AI developers is no longer being fought solely at the model layer. Increasingly, it is a contest over who can secure sufficient GPU capacity, data-center power, and networking bandwidth to train and deploy increasingly larger systems.

That is precisely where CoreWeave has carved out its niche. The company, often described as a “neocloud”, specializes in renting out high-performance AI infrastructure built primarily around Nvidia’s advanced GPU architecture.

Unlike traditional hyperscalers such as Amazon Web Services, Microsoft Azure, and Google Cloud, CoreWeave’s business is singularly focused on GPU-intensive AI workloads.

With Anthropic joining its customer roster, CoreWeave said nine of the ten leading AI model providers now use its platform, a striking validation of its infrastructure strategy.

The agreement reflects Anthropic’s aggressive push to lock in long-term compute access amid intensifying competition with OpenAI, Google, and Meta. The startup has been moving quickly to diversify its sources of infrastructure.

Earlier this week, Anthropic signed a long-term capacity arrangement involving Broadcom and Google, securing access to roughly 3.5 gigawatts of AI compute capacity beginning in 2027. Reuters also reported that the company is exploring the possibility of designing its own chips, a move that would mirror the broader industry trend toward vertical integration in AI hardware.

Together, these developments underscore a key reality of the AI boom: compute scarcity remains one of the industry’s most consequential bottlenecks. Practically, frontier labs are beginning to treat compute contracts much like airlines hedge jet fuel or manufacturers secure long-term semiconductor supply.

Any disruption in access to GPUs can slow training cycles, delay product rollouts, and erode competitive advantage.

One of the market’s longstanding concerns has been customer concentration, making the Anthropic deal a major financial win for CoreWeave. Last year, Microsoft accounted for roughly 67% of the company’s revenue, leaving investors concerned about overdependence on a single client.

This new agreement materially improves that picture. CoreWeave now has a diversified roster of blue-chip AI clients, including Microsoft, OpenAI, Meta, and Anthropic, which reduces concentration risk and strengthens revenue visibility.

That diversification matters because CoreWeave’s business is extremely capital-intensive. The company is pouring billions into data-center buildouts, GPU procurement, cooling systems, and power infrastructure.

Recent disclosures show it has been aggressively raising capital to keep pace with demand, including debt offerings and a major $8.5 billion loan facility to fund expansion. This has led some investors to question whether growth is being bought at the expense of balance-sheet strength.

Friday’s rally suggests the market is increasingly willing to look past those concerns, provided the company continues to convert capacity investments into long-duration contracts.

In the space of months, CoreWeave has secured an $11.9 billion agreement with OpenAI, a $21 billion expansion with Meta, and now a multi-year arrangement with Anthropic. That sequence effectively positions CoreWeave as one of the principal infrastructure beneficiaries of the AI supercycle.

The deal also comes with a broader industry implication. The fact that model developers are increasingly outsourcing vast chunks of compute capacity to specialized providers suggests that the AI economy is evolving into a layered ecosystem.

At the top sit the model builders. Beneath them are the compute suppliers, chipmakers, and data-center operators. In this hierarchy, CoreWeave is emerging as one of the most important infrastructure intermediaries, effectively the backbone provider for the model wars.

That makes Friday’s move in the stock more than a reaction to a single contract. Analysts see it as a market verdict on CoreWeave’s emergence as a strategic utility for the AI age. If demand for Claude and other frontier models continues to accelerate, this deal may prove to be another major step in cementing CoreWeave’s role as one of the industry’s most indispensable suppliers.

Anthropic’s beta launch of Claude for Word marks a fresh escalation in the enterprise AI battle, moving Claude deeper into the heart of daily office workflows and directly challenging Microsoft’s dominance in document software, legal tech, and knowledge work automation.

Anthropic has opened a new front in the enterprise AI war with the beta launch of Claude for Word, a move that goes far beyond another productivity add-in and signals a direct challenge to Microsoft’s hold over workplace software.

By embedding Claude natively into Microsoft Word, the AI startup is making an aggressive push into one of the most entrenched enterprise ecosystems in the world: document creation, contract review, memo drafting, and collaborative editing.

The immediate target appears to be high-value professional work, particularly legal services, finance, and enterprise advisory functions.

Anthropic said the new add-in is “designed for professionals who work extensively with documents, particularly in legal review, financial memo drafting, and iterative editing.”

The move is seen as not a mass-market writing assistant aimed at casual users but a deliberate attempt to move Claude into specialized, workflow-intensive environments where document precision, auditability, and revision control are essential.

For the legal profession in particular, the product has been built around real-world review processes rather than generic text generation. Users can ask the model to interrogate a contract and receive answers with clickable section citations, a feature that closely mirrors how lawyers and deal teams work through agreements.

Anthropic’s own example prompts underscore this focus. They include: “Summarize the key commercial terms: parties, term, governing law, and anything off-market.”

Other prompts include: “Flag provisions that deviate from standard market position, ranked by severity,” and “What did the counterparty change, and which revisions are dealbreakers?”

These are not generic prompts. They map directly onto core tasks in M&A, financing documentation, commercial contracts, and litigation review. That makes the product potentially significant for law firms, in-house counsel, private equity shops, and investment banks, where document turnaround time directly affects deal velocity.

The tracked changes integration may be the most strategically important feature. Rather than generating disconnected text in a separate chat window, Claude works inside the document layer itself.

Every edit can be accepted or rejected as a formal revision, preserving Word’s native redlining workflow. This is crucial for enterprise adoption because it keeps humans firmly in the review loop while making AI output operationally usable.

Anthropic says the tool can “edit selected text while preserving surrounding styles, numbering, and formatting,” and that a “tracked changes mode” lets users accept or reject every edit as a revision.

That solves one of the major friction points in enterprise AI deployment: workflow disruption. Instead of forcing professionals to leave Word, copy content into a chatbot, then manually paste it back, Claude is now positioned directly where the work happens.

Strategically, this is a far more consequential move than it may first appear for competitive reasons.  Microsoft has long treated Word, Excel, and PowerPoint as the cornerstone of its software empire, and its Copilot strategy is built on AI-enhanced productivity across that stack.

Anthropic is now effectively inserting its own intelligence layer into Microsoft’s moat. This is especially notable given that Claude had already been pushed into Excel and PowerPoint earlier this year, expanding its presence across the Office ecosystem.

The broader implication is that Anthropic no longer wants to be viewed primarily as a developer-first AI company. For much of its commercial rise, Claude has been strongly associated with coding, developer tooling, and enterprise APIs.

This latest rollout signals a much broader ambition. The company is now explicitly targeting knowledge workers across legal, finance, HR, strategy, and executive teams. In effect, Anthropic is positioning Claude not as a chatbot, but as an enterprise operating layer.

This is where the competitive pressure on Microsoft becomes more interesting. While Microsoft still controls the software environment, Anthropic is competing at the intelligence layer. That means the battle is shifting from who owns the application to who owns the decision-making and drafting workflow inside the application.

This raises a larger question about the future of productivity suites for investors and enterprise software watchers. The next moat may no longer be the document container itself. It may be the AI system that understands context, interprets legal nuance, tracks revisions, and accelerates professional decision-making.

If that thesis proves correct, Anthropic’s Word integration could be an early sign that AI companies are beginning to decouple value creation from the traditional software platforms they sit on top of.

For now, availability remains limited to Team and Enterprise plans, which reinforces the premium, business-first positioning of the launch.

The bottom line, however, is that Anthropic is moving aggressively beyond developers and into the core workflows of white-collar enterprise work, and in doing so, it is challenging one of the most durable franchises in corporate software.

Michael Burry, the investor made famous by “The Big Short,” has delivered one of his most detailed and scathing critiques yet of the AI-driven stock market, arguing that headline earnings for many Nasdaq 100 companies are badly overstated because of the way they account for stock-based compensation.

In a lengthy Substack post published this week, Burry said he spent weeks poring over more than 1,000 annual reports from Nasdaq 100 constituents going back a decade. His conclusion: Wall Street and the companies themselves are presenting an “earnings illusion” that makes valuations look far more reasonable than they really are.

Burry’s central argument is straightforward but uncomfortable for bulls. Companies and analysts routinely fail to fully account for the true economic cost of stock-based compensation (SBC). They should include not only the reported expense but also the cash spent on share buybacks to offset dilution, plus the net taxes triggered when those shares vest for employees.

Under current GAAP accounting, he calculates that Nasdaq 100 earnings are overstated by nearly 20%. That means if the index appears to trade at a price-to-earnings ratio of 25, the more accurate multiple is closer to 30 once SBC is properly adjusted for.

“Of every dollar of earnings per share that GAAP blesses, shareholders see only 83.49 cents of that dollar,” Burry wrote. “The wayward 16.51 cents wave crudely at GAAP and thumb their noses at shareholders on their way to employees’ pockets.”

He went even further on forward estimates. Wall Street’s consensus projections, he says, are running 42% higher than what he calls “true owners’ earnings” after full SBC adjustments.

Over the ten years ending in fiscal 2025, the 97 primary Nasdaq 100 companies reported a cumulative $4.9 trillion in GAAP net income. Analysts, often adding back SBC, pegged that figure at $5.8 trillion. Burry’s adjusted “true owners’ earnings” came in at just $4.1 trillion.

The $1.7 trillion difference, he argues, represents pure illusion — “the difference between what shareholders really owned of corporate earnings and what Wall Street and media reported.”

“Wall Street over the last 10 years guided investors to 42% more earnings than ever actually existed,” he added.

Burry singled out Meta Platforms as a clear example. He estimates the company has overstated owners’ earnings by about 20%. While Meta may look like it trades at 19 times forward earnings on the surface, the real multiple rises to 24 once SBC is properly factored in. If shareholders ultimately receive only about 83% of reported income, that multiple stretches closer to 28.

He reserved especially sharp words for Tesla, noting that the electric vehicle maker’s heavy use of stock-based compensation is so large that removing it from his dataset drops the overall Nasdaq 100 overstatement from 20% to 12.5%. Burry also took aim at Tesla’s massive $1 trillion pay package for CEO Elon Musk, calling it “such a beastly mass [that] would dwarf everything in my data set, even Tesla’s own epic deadweight.”

Other names Burry called out for particularly aggressive SBC practices include Datadog, Workday, Axon, Shopify, Palantir, Marvell, CrowdStrike, and Zscaler. He described the group collectively as “from an owners’ earnings’ perspective, a cesspool of shareholder disregard.”

Burry’s critique lands at a moment when many high-flying tech stocks are trading at punchy valuations fueled by the AI boom. The Nasdaq has repeatedly hit record highs, and forward multiples for the Magnificent Seven and other AI leaders have expanded significantly. His analysis suggests investors may be paying even richer prices than they realize once the full cost of keeping talent happy with equity grants is stripped out.

The veteran investor, who built his reputation calling the housing bubble and has a long history of issuing cryptic but pointed warnings, has been vocal about overvaluation risks in tech for some time. This latest deep dive, backed by a decade of granular data, is his most systematic attack yet on what he sees as a systemic flaw in how corporate America and Wall Street measure profitability in the age of stock-heavy compensation.

In sum, Burry’s message is: beneath the glittering GAAP numbers and rosy analyst forecasts lies a quieter transfer of wealth from shareholders to employees that is far larger than most realize — and one that could make today’s seemingly elevated valuations look even more expensive in hindsight.

While it’s not clear whether his warning will prove prescient or simply become another voice of caution in a momentum-driven market, it has been loudly delivered like every other.