In a note this week, analysts looked at the top five publicly traded AI hyperscalers: Meta, Alphabet, Microsoft, Amazon and Oracle.

BofA pointed out that while the companies can fund their near-term capital expenditures with cash, they are tapping debt markets for balance-sheet flexibility and better cost of capital. Last month alone, Meta, Alphabet, and Amazon raised tens of billions of dollars in the bond market.

Operating cash flow for the big five hyperscalers is expected to hit $577 billion this year from $378 billion in 2023, while debt should climb from $356 billion to $433 billion.

That means their overall debt burden is actually getting lighter as the debt-to-cash ratio should dip from 0.94 to 0.75.

“Given the hyperscalers’ historically conservative capital allocation and balance sheet policies, elevated debt issuance is possible, as evident by the recent bond deals from Meta, Alphabet and Amazon,” BofA said.

And plenty of additional cash is on the way. By 2029, operating cash flow is seen jumping 95% to $1.1 trillion, while capex is forecast to grow at a much slower pace of 58% to $632 billion.

But then there’s Oracle. Unlike the other AI hyperscalers, it will have negative free cash flow until 2029, meaning its capex will exceed cash from operations, according to BofA. As a result, it doesn’t have much capacity to take on more debt.

Indeed, fears about Oracle’s debt binge have rattled the overall AI stock trade as the company isn’t a cash machine like its AI peers.

Recent earnings guidance was also weak, and the company raised its forecast for fiscal 2026 capex by another $15 billion. In addition, surging lease obligations have spooked Wall Street.

A Financial Times report on Wednesday that said alternative investments firm Blue Owl didn’t team up with Oracle on a data center after all piled on more concerns. Shares fell on the news, though the company’s development partner, Related Digital, said Blue Owl was outbid on the project and didn’t back out of it.

But even though debt may not pose a limit on hyperscalers’ ambitions, they still face physical limits, namely in building enough infrastructure fast enough to meet demand.

Data-center researcher Jonathan Koomey told Fortune’s Eva Roytburg that capital can be deployed instantly, but the equipment that capital must buy cannot. Tmelines for turbines, transformers, specialized cooling systems, and high-voltage gear have stretched into years, he explained.

“This happens every time there’s a massive shift in investment,” Koomey added. “Eventually manufacturers catch up, but not right away. Reality intervenes.”

I’ve spent nearly five decades working across financial services, health care, and public  policy. I’ve served on presidential commissions, sat in closed-door briefings with lawmakers, and helped lead organizations working to meet the evolving needs of aging Americans. This crisis didn’t emerge overnight – we’ve seen it building for decades.  

For more than 30 years, commissions under Presidents George H.W. Bush, Bill Clinton,  George W. Bush, and Barack Obama all reached the same conclusion: our entitlement  programs were never built to handle a rapidly aging population. There were moments when  real reform seemed possible—when ideas were on the table and momentum was building. But again and again, the opportunities slipped by with inaction. 

Now we’re living with the consequences. By 2036, the population aged 85+ will more than  double. We’ll need nearly one million new assisted living units to meet demand, but we’re on pace to build only 40% of that.  

Most Americans still don’t understand how long-term care works, what it costs, or how to  prepare for it. And the reality is stark: home care now averages $77,792 per year, assisted living $70,800, and a private nursing home room more than $127,000—and those numbers are rising.  

Nearly 70% of Americans turning 65 will need some form of care, but more than 95% of baby boomers lack private insurance to pay for it. Most will rely on unpaid family caregivers or Medicaid, which only steps in after someone has spent down nearly everything they have.  

We are not prepared. Not families. Not the system. Not the economy. Not the country.  

Let me be blunt: the chance to enact sweeping reforms in time to help the baby boomers has passed.  

Structural reforms to Medicare or Medicaid are unlikely in today’s political climate, and  new federal rules are making it even harder to qualify for the latter. Both programs face  long-term sustainability challenges, but broad reform remains politically difficult—even as  insolvency looms. That’s not defeatism. It’s realism. 

So where does that leave us? 

Focus on the possible

We must focus on what’s still possible. And that begins with rethinking how care is delivered, how we define quality, and how we help people afford it.  

First, we need better planning tools. Today, most families make care decisions in a crisis—confused, overwhelmed, and without clear guidance. We must bring the same clarity to  aging that we do to financial planning: nurse-led evaluations, accessible education, and  unbiased support; not just product sales.  

Second, we need to raise the bar on quality. Too often, care is chosen based on  convenience or cost, not standards. Especially in home and community-based settings,  we must define what good, person-centered care looks like and build networks around  those expectations. This doesn’t require sweeping legislation—just transparency, data, and accountability. 

Third, we must confront affordability. The system punishes the middle class: too poor to  self-fund care, too rich to qualify for Medicaid. We need smarter contracting, vetted  provider networks, and eventually, portable, flexible insurance products that fill the gap.  Memory care, for instance, costs up to 30% more than traditional assisted living. Medicare fully covers just 20 days. Most people are left to cobble together care with out-of-pocket spending and fragile safety nets.  

Fourth, we must shore up the workforce delivering care. Care workers are leaving the  industry faster than we can replace them, driven by low pay, high demands, and little  support. Families are filling the gap, providing approximately $600 billion in unpaid care  each year while balancing jobs and other responsibilities. Nearly 60% of employees have  already provided care to a loved one, and most expect to in the future. Strengthening this  workforce—paid and unpaid—must be part of any serious path forward. 

We should also support bipartisan proposals like the WISH Act, which would create a national backstop for catastrophic long-term care events and their associated costs. At the state level, Washington’s WA Cares program offers a modest but meaningful  foundation. These models, paired with thoughtful private insurance solutions, point to a more realistic path forward.  

Moving beyond identifying the problem

We know what the problem is and who it’s hurting.  

What we need now is courage. Courage to act, to innovate, and to demand more from the system. Because the longer we wait, the more people fall through the cracks.  

The current system cannot stretch to catch everyone. It was never built to. And looking  away because the problem is complex, or politically inconvenient, is no longer acceptable. 

The baby boomers are aging into the final chapter of their lives. We owe it to them, and to  every generation that follows, to stop deferring action and start delivering solutions that meet the scale of the crisis. 

The companies declined to disclose financial terms of the transaction, but said it involves a mixture of cash and equity. They said Graphite will to continue operating as an independent product, but with deeper integration into Cursor’s code editing platform. The deal is expected to close in the coming weeks.

Cursor CEO, Michael Truell, told Fortune the acquisition addresses what he sees as an emerging bottleneck in software development.

“The way engineering teams review code is increasingly becoming a bottleneck to them moving even faster as AI has been deployed more broadly within engineering teams,” he said. “Over the past 2.5 years, Cursor has made it much faster to write production code. However, for most engineering teams, reviewing code looks the same as it did 3 years ago. It’s becoming a larger portion of people’s time as the time to write code shrinks. Graphite has done lots of work to improve the speed and accuracy of code review.”

AI code editors like Cursor help programmers while they’re writing code—making suggestions, explaining the function of a particular piece of code, and helping teams move around large projects faster. Graphite, used by companies like Shopify, Snowflake, and Figma, helps teams review changes and decide when code is ready to ship, after its written.

“We focused on the writing side of things. Graphite has focused on the review side of things. We think the two together can make something even better,” Truell said.

Graphite CEO Merrill Lutsky said that the two companies “have an almost identical vision for what the future of software development looks like.”

“Cursor has defined the new way to write code, and we’re defining how you review and merge it. Putting those together lets you build an end-to-end platform,” he told Fortune.

In the immediate term, both products will remain separate, with Graphite maintaining its independent brand. Throughout 2026, Truell said the companies plan to make it easier for developers’ code to connect with the review process, including smarter, more context-aware code review that adapts to how teams actually write code.

Lutsky said concerns about AI-generated code quality have been a major focus for Graphite. “We’ve invested deeply in ensuring that code written with the help of AI is safe and high quality,” he said. “Together with Cursor, we’re going to double down on that and help teams build secure, efficient, high-quality products.”

An end-to-end AI coding platform

The acquisition comes just one month after Cursor, which is valued at $29.3 billion valuation, announced it had reached $1 billion in annualized revenue. The company has seen a rapid rise since it was founded by a team of four MIT graduates in 2022. The company’s AI coding tool, which first launched in 2023, has seen major deployments at companies like Salesforce, which according to Truell said had seen a 30% uplift in engineering productivity from using Cursor.

Graphite is not Cursor’s first acquisition. The company bought AI coding assistant Supermaven in November 2024 and scooped up talent from enterprise startup Koala in July.

Graphite, which Lutsky co-founded nearly five years ago with Tomas Reimers and Greg Foster, raised $52 million in a Series B round in March 2025. The company told TechCrunch revenue grew 20x in 2024 without disclosing absolute figures, and expanded to serving tens of thousands of engineers at more than 500 companies, including customers such as Shopify, Snowflake, Figma, and Perplexity.

Lutsky said the deal offers Graphite the opportunity to build a more unified development platform. “We’ve long dreamed of connecting the surfaces where we create, collaborate on, and validate code changes,” he said, adding that the deal dramatically accelerates that timeline.

The AI coding market is booming

The AI coding market has exploded over the past two years as enterprises rush to adopt AI tools in hopes of productivity gains. The U.S. market for AI code tools was valued at $1.51 billion in 2024 and is expected to reach nearly $9 billion by 2032.

Big Tech companies including Microsoft and Google are automating large parts of their coding. According to Microsoft CEO Satya Nadella, as much as 30% of the code within the company’s repositories is now written by artificial intelligence while at least 25% of new Google code is generated by AI, according to CEO Sundar Pichai.

Companies are betting that AI coding tools can supercharge software engineers productivity, but early studies have been mixed. A July study by nonprofit research organization METR found that experienced developers using AI tools were actually 19% slower when using an AI coding assistant, even though they believed they were faster. Consulting firm Bain & Company also reported in September that real-world savings from AI coding have been “unremarkable.”

Nevertheless, the deal positions Cursor more aggressively in an increasingly competitive market, with OpenAI, Anthropic, and GitHub Copilot among those vying for dominance in the space. Most of these tools, however, are built on top of the same underlying “foundation” AI models rather than developing their own. Cursor, for example, uses Anthropic’s Claude and allows users to choose models from other providers to power code generation.

While Graphite is also backed by Anthropic, Lutsky downplayed concerns about competing directly with large model providers. “The larger base-model companies are trying to compete across many different verticals,” he said. “Cursor is solely focused on how engineers build with AI, and that focus really sets them apart.”

Truell also brushed off the threat from major AI labs. “Our approach here is to use a combination of the best technology that partners have to offer and then technology that we develop ourselves,” he said. The company has focused on cherry-picking the best available models, supplementing them with proprietary ones, and wrapping everything in what it argues is a superior user interface.

As for the next year, Truell said the company currently has no additional deals planned, with Cursor focused on building out product features rather than eyeing an IPO.

“Our goals for the company are very ambitious over the course of the next decade,” he said. “We think that this is the decade in which coding will be automated, and the way in which professional teams build and deliver software will change across the entire software development life cycle.”