Afternoon Brief · October 29th 2026

Afternoon Brief, October 29th 2026

AWS grows at its fastest rate in nearly five years, and Amazon falls anyway. Intel beats and then guides down. Canada buys into Cohere at a record . Mistral posts the for its trillion-parameter model.

13 min read·Editorial by Vera Lindqvist

VLVera Lindqvist
Capability & Systems

AWS grows 39%, its fastest in nearly five years, and Amazon still falls after hours

AWS revenue rose 39% to $46.0 billion in the third quarter, up from 37% the quarter before, and the unit now runs at $184 billion a year. Its was $17.8 billion against $11.4 billion a year earlier. Andy Jassy said he expects AWS to grow more than 40% in both 2027 and 2028 A.

The rest of the report is why the shares fell more than 8% after the close. Earnings of $1.89 a share missed the roughly $1.97 analysts wanted, and advertising revenue of $19.6 billion came in below the $20.8 billion expected. Amazon spent $77.3 billion on property and equipment in the quarter, more than twice the $35.1 billion of a year earlier, and trailing is now an outflow of $42.4 billion. It still expects about $220 billion in cash for 2026 A·B.

The release puts numbers on how much of that demand is model traffic: Fortune 500 spending on Bedrock grew nearly sevenfold from a year earlier and volume nearly seventeenfold, with OpenAI's GPT-6 models and Anthropic's Claude 5 models both on the platform. A year ago Amazon's net income was flattered by $9.5 billion in gains on its Anthropic stake, which is why profit fell while operating income rose 61%. SourcesA B B

Intel's data-center sales rise 78%, then its forecast sends the stock down 16%

Intel's third-quarter revenue rose 21% to $16.5 billion, and its Data Center and AI unit grew 78% to $7.3 billion, which CFO David Zinsner called the strongest data-center growth in 15 years. Adjusted earnings of 52 cents a share beat the roughly 40 cents expected B.

The fourth-quarter guide is what moved the shares: revenue of $15 billion to $16 billion against $17.1 billion expected, adjusted earnings of 25 cents against 43 cents, and falling to 38%. Zinsner put about $600 million of lost revenue over two quarters on an explosion at Intel's Israel in early October, and said the PC slowdown will last at least through the first half of 2027. The stock fell more than 16% , after losing more than 20% in the past month B.

Lip-Bu Tan said the "immediate priority" is unlocking more supply of Xeon server chips, which workloads have put back in demand because CPUs coordinate the work do. Intel also signed a multiyear deal to put more Xeon capacity, including future Diamond Rapids chips, into Oracle's cloud. SourcesB B

Canada takes up to C$1.6 billion of Cohere's record round, at a C$35.7 billion valuation

Cohere announced the first close of a targeting up to C$7 billion at a C$35.7 billion valuation, which Aidan Gomez called the largest private raise in Canadian history. The Schwarz Group, Lidl's parent and Cohere's partner in its merger with Germany's Aleph Alpha, leads. The new investors are Bell Ventures, Bullhound Capital and the Government of Canada; Inovia, Nvidia and Prosperity7 return A.

Prime Minister Mark Carney put the government's stake at up to C$1.6 billion of equity, announced in Montreal on the first day of the EU-Canada summit. "It is in all our interests for Cohere to succeed, and now all Canadians will have a share in that success," he said. Cohere says annualized revenue grew 165% in the year to the end of the third quarter but gives no absolute figure, so the multiple on that valuation cannot be checked A·B.

The valuation is well above the roughly US$20 billion that Cohere was reported to be seeking in September. A government putting its own money into one company on top of a European retailer's is a different kind of backing from a venture round: the pitch is sovereignty, and the buyers are states. SourcesA B B

Canada will join Europe's supercomputer network, the EU says

Canada and the European Union plan to sign an agreement on Friday that brings Canada into Europe's network of 12 supercomputers, letting Canadian companies and researchers train AI models on them, Ursula von der Leyen said at the summit in Montreal. Carney opened the day at Mila, the Quebec AI institute B.

Von der Leyen framed the AI cooperation as a hedge against a few dominant American firms: "Decisions taken elsewhere can put our businesses and essential services at risk." The agreement is announced, not signed, and its terms (how much compute, who pays, which machines) were not published. SourcesB

Mistral posts the weights for its trillion-parameter Large 4

Mistral released the weights of Mistral Large 4, a model with about one trillion and roughly 52 billion active per token, on in FP8 and NVFP4 builds. Mistral previewed the model earlier this month. The license is a modified MIT: anyone may use, fine-tune and sell on top of it, but a company selling the model itself as a hosted service needs a separate commercial license A.

Mistral's own numbers put it at 48% on Artificial Analysis's Coding Agent Index, level with Zhipu's GLM-5.3 while using 47% fewer tokens per task, 41% on Terminal-Bench 4.0, and 51% on Artificial Analysis's cyber index. It is pitched at defenders: Mistral reports it refuses 96.6% of malicious cyber requests and markets it as "sovereign cyber defense" for governments that want to run it themselves. The costs $0.70 per million input tokens and $2.10 per million output A·B.

The comparison Mistral chose is a Chinese model, which is where the frontier has sat most of this year. SourcesA B B

OpenAI cancels the Pro cut due tomorrow and doubles Codex allowances instead

OpenAI reversed a reduction that would have halved Codex and ChatGPT Work usage for $200-a-month Pro subscribers starting Friday. Codex lead Thibault Sottiaux posted that OpenAI is "doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today" C.

OpenAI's help pages describe the change more narrowly than the post: the Work and Codex allowance doubles on Pro 100 and Pro 500, Pro 200 subscribers move to the higher allowance, and weekly GPT-6 Pro message limits in Chat stay where they were. Prices do not change C.

The cut was announced a month ago, with efficiency gains as the reason and a one-time credit grant to soften it. Reversing it the day before it took effect, in a week when Anthropic launched spending controls for companies and GitHub said it routes agent work to cheaper models, says coding-agent allowances are now something the labs compete on. SourcesC C

Deepgram buys Daily, the company behind the open-source voice framework Pipecat

Deepgram, which sells speech recognition and synthesis, agreed to buy Daily, maker of Pipecat, one of the most widely used open-source frameworks for building voice agents. Terms were not disclosed. Daily had raised more than $62 million, according to PitchBook, and the two companies were in the same Y Combinator batch in 2016 B.

Both say Pipecat stays open source and vendor-neutral, so developers can keep plugging in rival speech and language models. Deepgram is also launching the Real-Time Context Standard, an open protocol co-written with the Pipecat team for passing audio context and tone instructions between the parts of a voice agent. A speech vendor owning the most popular neutral framework is the arrangement to watch: the pledge is the only thing keeping it neutral. SourcesB

Replika's founder raises $55 million for Wabi at a $1.25 billion valuation

Wabi, Eugenia Kuyda's startup, raised $55 million at a $1.25 billion valuation in a round led by IVP, with Andreessen Horowitz, Forerunner and the Alexa Fund. Angels include Tobi Lütke, Marc Benioff and Marissa Mayer B.

Wabi started as a way to build small apps from a prompt and now calls itself a "personal agent": a messenger that builds an interface on the spot for a task such as a meal plan or a trivia game. TechCrunch gives no user, revenue or retention figures, so the valuation rests on the founder and the category. SourcesB

Solink raises more than $100 million to read business data out of existing security cameras

Solink, an Ottawa company, raised more than $100 million led by Vertu Capital, with Goldman Sachs Growth Equity, Golub Growth and Comcast Ventures, taking its total past $198 million. Its software runs vision-language models and AI agents on the cameras retailers, restaurants and warehouses already own, for loss prevention and for screening false alarms in security operations centers B.

The company says about 500,000 cameras at more than 35,000 locations in 65 countries feed its platform. That installed base is the asset: it sells AI as a software layer on hardware nobody has to replace. SourcesB

AGIBOT proposes "Worken," a unit for how much useful work a robot fleet does

AGIBOT, the Shanghai maker, used its Innovation Day to propose Worken, a measure of robot output: active robots, multiplied by hours worked autonomously, multiplied by output against a reference such as a human worker's acceptable rate on the same task. It also unveiled AHA3, its third-generation humanoid hardware platform, with liquid-cooled joints for continuous operation and three new A.

AGIBOT says its 20,000th robot came off the line in September, but it published no Worken figures for its own fleet. A metric that discounts idle and teleoperated hours is the right question for a sector that reports production counts; it becomes evidence only when somebody publishes the number. SourcesA

Opentrons raises $40 million to make its lab robots something AI can drive

Opentrons, which makes benchtop liquid-handling robots, raised a $40 million Series E led by Avego with BroadOak Capital Partners. It says more than 11,000 of its robots are in labs in over 80 countries, including at 18 of the 20 largest drugmakers B.

The money goes to software and data infrastructure as much as hardware: Opentrons now sells its systems as "built for AI to run experiments directly." Cheap, standardized lab robots are the missing half of the AI-for-science programs the labs keep announcing. SourcesB

SuperMe raises $20 million to build AI stand-ins for real experts

SuperMe, founded in 2024 by former Pinterest and Grubhub growth executive Casey Winters and engineer Ludo Antonov, raised $20 million led by Forerunner's Kirsten Green, with Greylock. A user asks a question and gets a group chat with AI profiles of the experts best placed to answer, each built from that person's public writing and video and from apps they connect, such as Google Drive and Notion B.

It also works through , so the profiles can be queried from inside Claude or ChatGPT. The company has about 13 employees, has disclosed no usage, and plans to cap free questions. SourcesB

Sources

59 citations · 15 primary · 38 · 6 weaker

Morning Brief, October 29th 2026