Data center bargaining power shifts: previously dominated by cloud providers, now "CoreWeave and others" are starting to get tough
The strong bargaining position of cloud computing giants in AI data center leasing negotiations is weakening. Cloud providers such as Microsoft, in their eagerness to bring Nvidia servers online, are granting operators like CoreWeave more negotiating leverage—Service Level Agreement (SLA) terms are becoming more flexible, payment demands are increasingly strict, and even Nvidia and AMD are stepping in to bid for providing credit guarantees to clients.
Cloud computing giants were once the dominant party in AI data center leasing negotiations. In the past, large cloud providers such as Microsoft and Google, backed by strong capital and top-tier credit ratings, held the upper hand in negotiations with emerging cloud service providers and data center operators like CoreWeave, Nebius, and Nscale.
However, according to a report from The Information on September 7, as cloud providers like Microsoft are eager to get Nvidia server racks up and running as quickly as possible, data center operators like CoreWeave are gaining greater bargaining power, tilting contract terms in favor of the supply side.
This shift is reshaping the entire benefit structure of the AI infrastructure industry—from the stringency of service level agreements to the pricing of power procurement and the involvement of chip manufacturers in negotiations, all the rules are being rewritten.
Bargaining Power of Data Center Operators Rises, and Payment Terms Tighten Too
Service level agreements (SLAs) are the main battleground of contention.
Previously, large cloud providers demanded server rack uptime to be close to 100%, while also imposing extremely strict requirements on data center temperature and humidity.
According to a data center executive, he has seen contract terms such as: If a single rack goes down due to power outage, overheating, or switch failure, the cloud provider can cancel six months of rent. If SLA breaches accumulate to a certain point, the cloud provider could even directly terminate the lease.
The executive noted that negotiating SLAs is essentially a trade-off between "optimal price" and "contract durability"—the stricter the terms, the higher the price but also the greater the risk. “If you can secure SLAs with lighter penalties, it’s worth accepting a slightly lower price.”
Currently, as operators’ bargaining power increases, these extreme terms are being gradually softened.
It’s not just SLAs; payment terms are also shifting in favor of operators.
According to the report citing a credit executive, he has witnessed cases where a customer leased only a small section of a large data center, but the contract stipulated that if the customer fails to make timely payments, they must cover the rent for the entire facility for a certain period.
Even data center owners admit this requirement is tough: “He said, ‘Listen, we know it’s outrageous … but we can do it.’”
Nvidia, AMD Enter the Fray: Chip Makers Become Credit Guarantors
Operators are getting more leverage with another factor: proactive involvement by the chip makers.
According to the report quoting another data center executive, to ensure data centers equipped with their chips are built smoothly, Nvidia and AMD at times compete for the same data center project, each vying to provide credit guarantees for the customers.
“Nvidia is more aggressive because its balance sheet is stronger,” said the source.
These arrangements are especially advantageous for data center developers: Nvidia and AMD are willing to provide credit support for leases for up to 15 years, while previously, the duration Nvidia disclosed for some emerging cloud providers was just six years.
Power Costs: Up to 400% Price Variance in One Month
Power is the single largest component of data center operating costs—sometimes accounting for over a fifth of total costs, sometimes even more.
The above-mentioned credit executive stated that he has observed electricity prices fluctuate by as much as 400% among different clients within the same month.
The paradox is that every customer believes they’ve secured a great deal, but in reality, the largest cloud providers often pay the highest power prices—because they can afford it.
The pressure from electricity shortages is also spreading. According to a previous report from The Information, Elon Musk predicts a significant power shortfall by 2027 and is planning to build his own turbine blade factory. Meanwhile, some data center developers are purchasing natural gas turbines at high premiums from dubious “shell companies,” further pushing up insurance and financing costs.
All these factors combined are making it extremely difficult to forecast data center costs.
According to the report, in recent years, the cost to construct a 1 GW data center has more than doubled. Should costs continue to rise, investor sentiment in many of the world’s largest listed companies will come under pressure.
The Fragmentation of Compute Power: Nvidia’s Alternative Game Plan
With supply tight for large-scale data centers, Nvidia CEO Jensen Huang proposed another path forward at the Equinix customer conference.
“Compute power is fragmenting,” he stated at the conference. “The AI world will fundamentally decentralize and become highly distributed.”
Nvidia, Equinix, and Together AI announced a joint plan at the conference: Together AI will purchase Nvidia hardware and deploy it within Equinix’s existing data centers to offer inference services for open-source AI models to small and medium-sized enterprises.
Huang described the architecture’s logic: latency-sensitive AI tasks run in Equinix data centers closest to users, while memory and inference tasks can be handled further afield.
Executives from Google, Cisco, and the emerging cloud service provider Lambda Labs expressed similar outlooks at the conference: AI inference is moving toward decentralization, and a distributed network formed by thousands of small facilities could become the mainstream model in the future.
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
The script at the end of 2018: The Federal Reserve continued to raise interest rates, US stocks plunged 20% from the September highs, forcing the Federal Reserve to "pivot" in January.
Goldman Sachs' head of trading believes the current US stock market is repeating the scenario of late 2018, with monetary tightening and equity market pressure. The Goldman Sachs Financial Conditions Index has tightened by about 8 points in two weeks, reaching the highest level in a year. The selloff has broadened from tech stocks to defensive sectors. Gold and equities are declining simultaneously, possibly indicating investors are selling assets to raise cash, making gold a key indicator for observing liquidity stress.

Goldman Sachs Trader “Revisits the 2018 Playbook”: Still Clearly Remembers “On December 24th, Clients Asked Why the Market Dropped So Much—And I Had No Idea”
Brian Garrett pointed out that the current adjustment in the US stock market is similar to the sharp decline in Q4 2018: the S&P 500 has already experienced 42% of the corresponding drop, and the VIX is approaching its peak just before Christmas. If history repeats itself, the bottom of this cycle may appear in January next year. However, the credit market and Federal Reserve policy have yet to provide clear signals of a bottom.
Anthropic (ANT.US) nearly “broke issue price,” but Claude knocks on the doors of “the highest peak in mathematics”! 17 million lines of Lean code used to write a candidate proof for the Riemann Hypothesis
On November 20th, Eastern Time, on its second day of trading, Anthropic's intraday price dropped to a low of $83.05, just 5 cents above the offering price of $83, and finally closed at $88.91, a decline of 8.50%. On the same day, the company's mathematician Levant Alperg announced on his personal website that he and Claude had found a candidate proof for the Riemann hypothesis, which has puzzled the mathematics community for 167 years, and released about 17 million lines of Lean formalized code.

As US stocks plummet, the parent company of GTA 6 rises against the market! First-day sales hit a record $3.79 billion; Wall Street shifts focus to next-generation GTA Online.
According to Rentrak data, the global box office of "Grand Theft Auto VI" has also surpassed that of "Avengers: Endgame," the highest-grossing film of all time, which has a global box office of approximately $2.94 billion.

