Microsoft's cloud growth stumbles

Welcome to Runtime! Today: how a narrow miss on cloud growth raises questions about Microsoft's AI spending, why Nvidia wants AI-curious businesses to use its new managed deployment services, and the latest funding rounds in enterprise tech.

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

Great expectations

No enterprise tech company has been more closely identified with the generative AI boom than Microsoft, which staked billions of dollars in late 2022 on the premise that AI will forever change the way businesses build and use software. Investors have been looking for signs of progress toward that goal ever since, and they didn't like what they saw from its fourth-quarter earnings results.

Microsoft shares fell more than 6% in after-hours trading Tuesday after Microsoft reported Azure growth of 29%, which fell short of Wall Street's expectations for 31% growth according to CNBC. Overall revenue and profits beat those expectations, but Azure growth is the most closely watched metric for assessing Microsoft's success in getting enterprise customers to buy into the glorious AI future they've been promised.

• Microsoft attributed 8 percentage points of that 29% growth to AI-related services, up a point from the contribution it made last quarter.
• However, Azure grew at 31% last quarter, which implies that non-AI-related Azure growth at Microsoft is slowing.
• Microsoft CFO Amy Hood attributed that slower growth in non-AI Azure business to "softness we saw in a few European geos" but also said demand for Microsoft's AI services outpaced supply, which limited the revenue upside.
• Both conditions are expected to continue during the rest of the year, and Hood forecast Azure growth of 28% to 29% in the current quarter.

Microsoft 365 is the other important product for assessing Microsoft's AI fortunes, thanks to the introduction of the 365 Copilot product last November, and the news there was a little better.

• The number of business customers using 365 Copilot grew 60% compared to last quarter and the number of customers that bought more than 10,000 seats doubled, CEO Satya Nadella said.
• "The most healthy sign to me is the fact that customers are coming back," Nadella said. "The same customers with whom we landed the seats are coming back and buying more seats."
• There definitely seems to be a learning curve for early customers of Microsoft 365 Copilot, who have struggled to feed it the right amount of data and use it across Microsoft products like Word and Excel, as Bloomberg reported last week.
• And it's not cheap: Adding Copilots to corporate 365 subscriptions can double the price per person per month, which means companies will need to see a long-term return on that investment to keep coming back.

The sell-off rebounded as Nadella and Hood ran through their talking points on the conference call, but the stock was still down almost 4% after the call ended. While almost every company in tech would trade places with Microsoft in a heartbeat and slightly disappoint investors by reporting net income of $22 billion, the amount of money it is spending on AI infrastructure makes those same investors more nitpicky.

• Microsoft spent $19 billion on capital expenditures in its fourth quarter, much higher than the $14 billion it spent during the third quarter and the same amount it spent during its entire 2020 fiscal year.
• And, as noted, Hood said it was still constrained on supply.
• All-up (that's Microsoft-ese for "in total"), it spent $55.7 billion on capital expenditures during the 12 months ending on June 30th, which is nearly as much as it spent in 2022 and 2023 combined.
• Hood said capital expenditures will increase during the current fiscal year.

It takes a long time for capital expenditures to pay off as revenue. As long as Microsoft continues to print money from its overall business, investor grumbling should be muted.

• But slowing cloud growth almost two years into the generative AI boom is not what most people had in mind.

A raft of NIMs

Nvidia's rise to the top of the AI boom has always had as much to do with its software strategy as its GPUs, which are still hot commodities. This week it rolled out several new additions to its NIM (Nvidia inference microservices) family of AI app containers that were designed to make it easier for businesses to deploy AI into production.

NIMs are a collection of packaged APIs that are part of Nvidia's AI Enterprise service and allow corporate applications to run inference workloads against popular AI models. They can be deployed to Kubernetes and coordinate the interactions between chips, AI models, and all the software needed to use AI without having to stitch together the pieces.

The new NIMs promise to let companies build their own somewhat-creepy "digital human" assistants and use RAG (retrieval-augmented generation), a technique designed to improve the accuracy of AI model results. However, they don't come cheap: The Nvidia AI Enterprise service costs $4,500 per GPU used per year, according to Bloomberg.

Enterprise funding

Altana raised $200 million in Series C funding that elevates the supply-chain software company to unicorn status.

Chainguard secured $140 million in Series C funding as it expands its security services for companies that want to identify risks in their software supply chains.

Credo AI landed $21 million in new funding to help companies introduce governance standards across their AI applications.

Lineaje raised $20 million in Series A funding for its own approach to software supply-chain security, which helps companies produce a software bill of materials.

Lakera scored $20 million in Series A funding to help businesses using LLMs defend against prompt-injection attacks.

Monto launched with $9 million in seed funding to bring its simplified business-to-business payments software to U.S. customers.

The Runtime roundup

Bausch Health is the latest Snowflake customer to confirm that sensitive data was stolen by hackers who bypassed weak account security controls, 404 Media reported.

Apple trained its newest LLMs on Google Cloud's TPU chips rather than Nvidia's GPUs, it revealed in a paper this week.

AMD reported data-center revenue growth doubled compared to last year, and raised its guidance for this year.

SAP shook up its leadership team, and Chief Marketing and Solutions Officer Julia White and Chief Revenue Officer Scott Russell are out according to CIO.

Clickhouse acquired PeerDB to add new Postgres capabilities to its data warehouse.

World records are falling this week, and not just in Paris: The Dark Angels ransomware group recently received a payment of $75 million, which appears to be a new high (or low).

Thanks for reading — see you Thursday!