Advanced Micro Devices will sell Anthropic tens of billions of dollars' worth of AI servers and invest as much as $5 billion in the Claude maker, a deal that tightens the grip of chip and cloud power over the people building and buying AI tools. Anthropic will buy up to two gigawatts of AMD's latest-generation Instinct MI450 chips, starting in the first half of 2027, while AMD's investment will depend on hitting deployment milestones. AMD executives have said 1 gigawatt of computing power, enough for roughly 750,000 U.S. homes, can cost double-digit billions of dollars. That’s the scale of the machine. Massive, expensive, and built to serve a race run by giants.
Who Gets the Compute
Tom Brown, co-founder and compute head at Anthropic, put the dependency in plain language: "Access to compute is central to keeping Claude at the frontier and meeting demand from our customers." Anthropic has moved aggressively in recent months to overcome capacity constraints for its services. In May, it agreed to rent the full computing power of SpaceX's Colossus 1 facility in Memphis, which houses more than 220,000 Nvidia processors and gave it 300 megawatts of new capacity. Meta Platforms is also in talks to lease computing power to Anthropic in a potential deal worth up to $10 billion over two years, according to a source familiar with the matter last week. The startup’s growth, in other words, depends on access to industrial-scale infrastructure controlled by a handful of firms.
Under the latest deal, Anthropic would use some of the AMD chips at its own data centers and lease additional capacity through cloud providers and new AI cloud companies, according to the Wall Street Journal, which first reported the news. AMD was in talks to provide a financial backstop for Anthropic's future data-center leases, the report said. Shares of the Santa Clara, California-based company rose 2.4%, and the stock has more than doubled in value so far this year. In October, AMD announced a multi-year deal with OpenAI that would also bring in tens of billions of dollars in annual revenue while giving the ChatGPT creator the option to buy up to roughly 10% of the chipmaker. The circle keeps tightening.
The Circular Deal Economy
The announcement marks the latest so-called circular deal in the AI industry, where chipmakers invest in AI firms that are among their biggest customers. Nvidia has been in talks to invest $30 billion in ChatGPT creator OpenAI. That arrangement, and the AMD-Anthropic deal, show how the same corporate players keep financing one another while ordinary people get told this is innovation. The money moves upward. The control stays concentrated.
U.S. hyperscalers are starting to show returns on their artificial intelligence investments, but the rising cost of the buildout is taking a bite out of their free cash flow, and investors are noticing. At their current trajectory, the so-called hyperscalers -- Microsoft, Alphabet, Amazon, Meta Platforms and Oracle -- are expected to spend more combined on capital expenditures than they generate in free cash flow by 2027, according to a Reuters analysis of LSEG consensus estimates. The data shows the companies will generate about $340 billion more in annual operating cash flow in 2027 than in 2025, but capex is expected to rise by roughly $534 billion, equivalent to about $1.57 of additional investment for every $1 of additional cash flow. The buildout eats cash. The market wants growth. The workers and users get the bill in the form of more infrastructure, more extraction, more pressure.
When those companies report earnings, beginning with Alphabet on Wednesday, investors will be looking for signs that the rapid growth in cloud and AI revenue can keep pace with the expected spending surge. Hyperscalers led the market rally since the AI buildout began, surging on the promise of future growth. Over the last year, all but one -- Alphabet -- have trailed the S&P 500. Shay Boloor, chief market strategist at Futurum Equities, said, "Investors are underestimating how fundamentally AI is changing the Big Tech business model." He said, "These companies were historically valued as asset-light platforms because revenue could scale much faster than capital requirements, but AI is pushing them toward a hybrid model where software, advertising and cloud economics increasingly depend on enormous physical infrastructure spending."
Who Pays for the Machine
The capex estimates cover all spending, not just AI-related expenditures, because companies do not consistently disclose AI-specific investment. Spending on data centers, servers, networking equipment and other cloud infrastructure is largely driven by AI demand, executives have said. The spending outlook is also subject to change. Current-year consensus estimates for the five companies' capex have risen from about $485 billion in January to around $730 billion in July, according to LSEG. There are some signs that AI spending is paying off. Microsoft has said its AI business has surpassed a $37 billion annual revenue run rate, while Amazon reported 28% growth at its AWS unit in its first quarter.
Still, for investors, the worry is whether the AI-related cash generation falters while spending persists. Microsoft reported $35.8 billion in operating cash flow in its fiscal second quarter while recording $37.5 billion of capital expenditures, including finance leases. David Russell, global head of market strategy at TradeStation, said, "Earnings growth may not be enough to justify investment if capex is depleting cash. Companies exist to make money, not spend money." Amazon said trailing 12-month operating cash flow rose 30% to $148.5 billion in the first quarter, but free cash flow fell to $1.2 billion.
Investors seem most alarmed about Oracle, whose shares have lost 36% so far this year as its free cash flow has turned negative. Its capex as a percentage of operating cash flow has steadily increased, and it plans to raise $45 billion to $50 billion through debt and equity to fund cloud infrastructure expansion. Oracle's capex as a percentage of operating cash flow rose from 47% in fiscal 2022 to 174% for fiscal 2026, which ended in May, according to LSEG data. Capex came to $55.7 billion in its most recent fiscal year, compared with operating cash flow of $32 billion. The other companies so far have been able to return cash to shareholders. Microsoft, Alphabet and Meta still generated enough free cash flow to cover dividends and buybacks in their latest fiscal years, according to SEC filings. Buybacks could be at risk if spending remains elevated and AI monetisation takes longer than expected. Freddy Lavric, senior trader at Winthrop Capital Management, said, "Over the next two to three years, companies need to show that AI is driving incremental revenue, expanding margins and improving cash flow." He said, "If those financial benefits aren't becoming evident by then, the market will start questioning whether the investment cycle has gone too far."