Good day fellow AI and Tech friends!

AI’s center of gravity moved toward the machinery that makes it useful and expensive. Infrastructure spending stayed hot, operational deployments widened, and publishers escalated their fight over traffic lost to AI summaries.

1. CoreWeave and Super Micro say the AI buildout is still accelerating

What happened: CoreWeave reported second-quarter revenue of $2.58 billion, more than double a year earlier, and raised its 2026 capital-spending forecast to $35–$39 billion. Its revenue backlog reached roughly $104 billion, before more than $25 billion in early-third-quarter customer commitments. Separately, Super Micro forecast fiscal 2027 revenue of $65–$72 billion, well above the $52.5 billion analyst average cited by Reuters.

Why it matters: These are two different businesses telling the same story: demand for AI compute remains strong enough to justify enormous spending. CoreWeave’s $626 million quarterly net loss is the useful reality check. Selling GPUs by the cloudful is lucrative; financing and powering them is not a hobby for the faint of balance sheet.

What to watch: Power, cooling and networking delays—not model quality—may be the near-term bottleneck. Watch whether backlog converts into delivered capacity without debt costs swallowing the gains. Sources: CoreWeave results, Reuters on Super Micro.

2. IBM and Together AI put $240 million behind open-model inference

What happened: IBM and Together AI signed a multi-year $240 million agreement for a large AI inference cluster on IBM Cloud. The planned U.S. deployment will use Nvidia HGX B300 systems and Spectrum-X networking, with availability expected in the first quarter of 2027. Reuters reported an initial configuration of about 2,000 Blackwell 300 chips.

Why it matters: Training produces the model; inference produces the bill, every time someone uses it. The deal is evidence that open-weight models are becoming an enterprise infrastructure market, not merely files developers download after midnight. It also gives IBM a clearer role in a cloud segment dominated by larger hyperscalers.

What to watch: The commercial test is “token economics”: whether open models on dedicated hardware can beat closed-model APIs on cost, reliability and security. Source: IBM announcement.

3. Nvidia’s reported Nemotron 4 plan pushes beyond chips

What happened: Reuters, citing The Information, reported that Nvidia is developing the Nemotron 4 model family. The largest version is expected to have at least one trillion parameters and could arrive as early as late fall, but final training is unfinished and no release date has been set. Nvidia confirmed it is investing in Nemotron, not the reported specifications. The company also released Nemotron 3.5 Lightning and the open-source NeMo Switchyard routing library on Tuesday.

Why it matters: Nvidia wants to shape the software layer that drives demand for its hardware. A credible open-weight family would also give enterprises and governments another U.S.-developed alternative as Chinese open models improve. For context, see our earlier comparison of the open-model field.

What to watch: Treat the trillion-parameter figure and fall timing as reported plans, not settled facts. The license, training data disclosure, cybersecurity safeguards and real benchmark results will matter more than the parameter count. Source: Reuters.

4. China tests AI weather models against a real typhoon

What happened: Chinese systems including Fengwu, Huawei’s Pangu and Fudan University’s Fuxi worked alongside conventional forecasting during Typhoon Dolphin. Researchers say these models can produce forecasts far faster than physics-based supercomputer systems while matching or beating them on some measures. One company involved with Fengwu said it predicted the storm’s landfall five days ahead to within 30 minutes and 30 kilometers.

Why it matters: Weather forecasting is a high-value AI use case with measurable outcomes: better track predictions can improve evacuations, transport planning and flood preparation. It is also a useful antidote to benchmark theater. The atmosphere, inconveniently, does not optimize for leaderboard scores.

What to watch: The quoted landfall result is a developer claim. AI models still lag conventional systems on storm intensity and remain unproven for longer-range climate events, so hybrid forecasting is likely to continue. Source: Reuters.

5. French publishers escalate the fight over Google’s AI summaries

What happened: The Alliance of General Information Press asked France’s competition authority to act against Google’s AI-generated article summaries. The publisher group argues that Google rolled out the feature without authorization or dedicated payment. It cited French communications regulator Arcom’s estimate that AI summaries have reduced traffic to publisher sites by 33%–38%. Google had not responded to Reuters at publication time.

Why it matters: This dispute goes beyond copyright. It asks whether search platforms may use publishers’ reporting to answer a query so completely that the source loses the visit—and then call the arrangement mutually beneficial.

What to watch: France previously fined Google €500 million in a news-content dispute. A new competition ruling could influence negotiations elsewhere in Europe. The traffic figure is a regulator estimate cited by the publishers, not an independently audited number in the Reuters report. Source: Reuters.

6. Ryanair brings Gemini into crew and maintenance decisions

What happened: Ryanair announced a five-year Google Cloud agreement covering 35,000 employees. The airline says it will use Gemini Enterprise to build custom agents for decision automation, crew scheduling and disruption management, plus DeepMind models including AlphaEvolve and WeatherNext for fleet operations and maintenance scheduling. Financial terms were not disclosed.

Why it matters: This is the kind of deployment that will determine whether enterprise agents graduate from slide decks. Airline operations are constrained, time-sensitive and expensive when they go wrong. Ryanair is keeping AWS in a dual-cloud setup, which is a practical vote for resilience over vendor romance.

What to watch: The announcement describes intended uses, not measured results. Look for evidence on delay reduction, scheduling accuracy, human oversight and how the airline handles agent errors in safety-adjacent workflows. Source: Reuters.

The one thing to remember

Yesterday’s signal was not one spectacular model release. It was the widening industrial footprint around AI: capital, chips, inference, operations, forecasting and content economics. The technology is becoming infrastructure—and infrastructure is where promises meet power bills, regulators and weather.