How the U.K. could shake up the cloud

Welcome to Runtime! Today: Runtime read the 254-page Ofcom report on cloud computing competition so you don't have to, Microsoft's GitHub is losing a lot of money serving up code suggestions, and the latest funding for enterprise tech startups.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

The once and future cloud

Last week Ofcom, the U.K.'s regulatory agency for telecommunications, recommended that the country's competition authorities take a closer look at established cloud infrastructure business practices, citing concerns about the ability of customers to switch providers and employ multicloud strategies. In a 254-page report published on its website, Ofcom laid out its case after talking to several major providers and ISVs, including AWS and Microsoft, as well as conducting "50 one-hour discussions and over 1,000 survey interviews with U.K. decision-makers in U.K. businesses that used, or were considering using, IaaS and/or PaaS services."

As novellas about cloud computing go, the report is well-written and grounded (Protocol Enterprise was cited at least twice) in its understanding of the market forces that have grown up around cloud computing, but prone to some magical thinking. It singles out three main problems that it believes restrict the choices of cloud customers, and those concerns will be familiar to anyone who has followed this space in recent years.

• Most cloud customers are charged what Ofcom considers to be excessive egress fees to move data out of cloud data centers, and, unsurprisingly, "our customer research found that 78% of respondents thought egress fees should be reduced or removed."
• Most cloud providers implement similar computing services in slightly different ways for a variety of reasons, and Ofcom is "concerned that some of the barriers which arise from technical differentiation are not justified."
• And it took aim at the so-called "committed spend" contract strategy employed by many cloud providers, especially AWS, to lock in discounts for computing services if customers agree to spend a certain amount over a certain amount of time: "The structure of these discounts acts as a barrier to muticloud by encouraging larger customers to use a single hyperscaler for all or most of their cloud needs."

One aspect of the report that felt a little weird was how it conflated customers using basic cloud services with those using managed cloud services when discussing concerns about switching costs and multicloud computing.

• The best way to quickly switch between cloud providers or implement a multicloud environment is to use the most basic infrastructure services provided by the clouds, such as compute and storage, and design your own custom tooling to fill in the gaps, which is what Target did.
• But companies that want to use managed versions of those infrastructure services make that decision because they want to spend more time worrying about their core businesses than their infrastructure, which is what Spotify did.
• Managed services are a premium product and cloud customers are not required to use them; it doesn't seem that controversial that the use of those managed services would be tied to the vendor that built those services.

Ofcom does not impose interventions, but it does recommend them. Some of those recommendations are straightforward.

• In its view, cloud providers should be more transparent about compatibility issues between their own products, competitors' products, and open-source projects.
• It also explored, but wisely dismissed, suggestions that U.K. regulators impose standards for building cloud infrastructure services, which would be a nightmare in practice.
• Ofcom pointed out that groups like the Bandwidth Alliance promise low-cost or free outgoing data transfers between members, suggesting that doing away with egress fees entirely would be easier than having regulators set a price but also acknowledging that cloud providers have constructed private networks at substantial cost.

Others are sure to make cloud computing in the U.K. a lot more complicated and would rely heavily on regulators making good, informed decisions when writing the rules, which…

• Ofcon suggested that it might make sense to force cloud providers to directly connect their data centers to boost interoperability, similar to the arrangement that Microsoft and Oracle have reached in the U.S., but acknowledged "we expect such intervention to require a complex design, including which cloud providers it applies to, defining an availability zone and setting out requirements on the type of connectivity."
• It also wants regulators to consider forcing cloud providers to open their APIs up to third-party software, making it easier to run Azure Kubernetes Service, for example, on AWS's EC2.
• Even assuming there are people who want to do that, Ofcom noted that an unnamed cloud provider strongly objected to that proposal, citing "serious detrimental effects on the integrity, security and quality of services provided."

Now it's up to the U.K.'s Competition and Markets Authority, which has the ability to mandate certain practices in order to do business in the U.K. and promised to complete its investigation by April 2025.

We'll make it up on volume

The unit economics of the early generative AI boom are seriously out of whack, the Wall Street Journal reported Monday. Microsoft's GitHub is losing about $20 a user on its Copilot service each month, according to its sources, and it is losing a great deal per month more on the power users.

It's no secret that training anything on OpenAI's GPT-4 model is extremely expensive, and that computing costs will need to come down or application prices will need to go way up to make this hype cycle eventually pay off. But it's hard to see the computing costs coming down in the near future, given the supply constraints around Nvidia's AI chips, the data-center infrastructure upgrades needed to support this new style of computing, and the yet-unrealized attempts to use smaller models for generative AI tasks. 

For now, Microsoft is sticking with a flat per-user fee for its first AI services, with plans to charge Microsoft 365 Copilot customers $30 per user per month when it arrives in November. Mid-tier Microsoft 365 customers are looking at doubling their monthly cost to use the new AI features; how much more will they want to pay?

Enterprise funding

Gutsy raised an eye-popping $51 million seed round and launched its process-mining cybersecurity service, developed by the team that built Twistlock.

Observe scored $50 million in Series A funding to support the launch of a new version of its observability service that includes generative AI features.

Opsera landed $12 million in new funding to further the development of its DevOps platform service.

The Runtime roundup

The biggest DDoS attacks yet recorded hit AWS, Microsoft, Google, and Cloudflare over the last two months using a flaw in the HTTP/2 protocol.

AMD acquired Nod.ai, a startup working on developing software that makes it easier to implement AI models.

Google and Microsoft separately released new AI tools for the health-care companies looking for better ways to organize and manage patient information.

Elon Musk shut down a Twitter disinformation service months before Hamas' attack on Israel last weekend in part because he wanted to reduce his Google Cloud bill, according to The Information.

Thanks for reading — see you Thursday!