Tech Current Daily Brief — August 12, 2026

The clearest signal from the last 24 hours is that the AI buildout is becoming a full-stack infrastructure story, not just a model race. CoreWeave’s latest quarter shows demand still pulling enormous amounts of capital into GPU cloud capacity, while Lumentum’s results show that optical interconnects are becoming a meaningful beneficiary of the same expansion. IBM is locking in a dedicated inference cluster for Together AI, and Google is extending agentic development from code generation into real-device testing. Outside AI, Joby’s $500 million purchase of Resonant Sciences shows another capital-intensive frontier-tech company broadening into defense. The common thread is execution: capacity, networking, testing and adjacent revenue are moving to the foreground.

Today’s 5 Top Stories

CoreWeave posts $2.58 billion in Q2 revenue and raises the scale of its AI buildout

CoreWeave reported second-quarter revenue of about $2.58 billion on Tuesday, more than double the year-earlier level and slightly above Wall Street expectations. The AI cloud provider also reported a smaller adjusted net loss than analysts expected. More important for the infrastructure cycle, CoreWeave ended the quarter with roughly $104 billion of revenue backlog and spent about $9.4 billion on capital expenditures; management also lifted its full-year capital-spending outlook to $35 billion–$39 billion. Those numbers underline the unusual economics of the current AI cloud market: demand can be booked far ahead, but converting that demand into revenue requires data centers, power, GPUs and financing to arrive on schedule. The company’s after-hours share move reflected relief that revenue growth and operating leverage are keeping pace with the spending.

Why it matters: CoreWeave’s constraint is increasingly execution and financing, not finding customers; the key question is how efficiently backlog becomes powered, billable capacity.

Lumentum’s quarterly revenue reaches $1.01 billion as AI optics demand accelerates

Lumentum reported fiscal fourth-quarter revenue of roughly $1.01 billion, up 109% year over year, with adjusted earnings of $3.23 a share, up 267%. The optical-components supplier also forecast fiscal first-quarter revenue of $1.225 billion to $1.275 billion and adjusted earnings of $4.05 to $4.35 a share, both ahead of prevailing analyst expectations. Lumentum has tied its recent growth to cloud and AI data-center demand, including newer 1.6-terabit transceivers and high-power lasers used in advanced optical systems. The important point is not the share-price reaction; it is where AI infrastructure spending is showing up. As accelerator clusters become larger, more of the system budget moves into connecting GPUs within and between racks and data-center campuses.

Why it matters: AI infrastructure bottlenecks are broadening beyond accelerator supply, making optical bandwidth and photonics capacity increasingly important to system economics.

IBM and Together AI sign a $240 million deal for a B300 inference cluster

IBM and Together AI entered a multiyear $240 million agreement to add a large Nvidia HGX B300 cluster to IBM Cloud, according to an announcement Tuesday reported by Barron’s. The cluster is expected to launch in early 2027, with Together AI using the capacity for open-model inference. The deal is notable because it is explicitly aimed at inference rather than frontier-model training, and because it puts a specialist AI platform on infrastructure supplied by a long-established enterprise cloud provider. Together AI has been expanding its pool of dedicated Blackwell-class capacity as it tries to serve production workloads built on open and custom models. For IBM, the contract creates a concrete AI workload on its cloud rather than relying only on software and consulting exposure to the AI cycle.

Opportunity signal: Long-duration inference contracts are creating room for cloud operators, capacity brokers and software layers that can improve utilization, routing and cost control across heterogeneous GPU fleets.

Google launches Developer Device Platform for agentic mobile testing

Google announced the public preview of Developer Device Platform, or DDP, on August 11, with availability to Google Cloud users beginning August 12. DDP provides on-demand access to physical devices and high-concurrency virtual emulators, extending the company’s Firebase Test Lab lineage into a platform designed for coding agents as well as human developers. Its Device Streaming API supports interactive access to an emulator or physical device, while Device Run can execute tests in parallel across hundreds of devices. Google says an upcoming DDP agent skill will let coding agents execute multi-step user journeys, spot visual defects, inspect on-device chip performance and validate hardware-specific fixes. Public-preview usage is billed by active testing time.

Why it matters: If coding agents are expected to ship production mobile software, they need access to execution environments and hardware feedback—not just source code, terminals and synthetic tests.

Joby agrees to buy Resonant Sciences for $500 million and create a dedicated defense division

Joby Aviation announced Tuesday that it will acquire Dayton, Ohio-based Resonant Sciences in a $500 million transaction, funded with $450 million in cash and $50 million in Joby shares. The deal is expected to close in the first half of 2027. Resonant has roughly 250 employees and works on sensing, antennas, secure communications, signal processing and low-observability technologies; its products are qualified across more than 20 commercial and defense airframes, according to reporting on the deal. Joby plans to keep its commercial air-taxi program separate while making Resonant the core of a dedicated defense business. That gives Joby a larger base of cleared personnel and mission-systems expertise alongside its propulsion and autonomy work.

What to watch: The strategic test is whether Resonant produces durable defense programs and integration advantages without diverting capital or engineering attention from Joby’s commercial certification and launch plans.

Data & Market Pulse

The most useful numbers today are infrastructure numbers. CoreWeave’s roughly $104 billion backlog is more than 40 times its latest quarterly revenue, but its approximately $9.4 billion of Q2 capital spending shows the cost of turning those contracts into live capacity. Lumentum’s revenue more than doubled to about $1.01 billion, then guided to as much as $1.275 billion for the next quarter, evidence that AI spending is propagating into optical components rather than remaining concentrated in accelerators. IBM’s $240 million Together AI contract is smaller than the hyperscaler capex programs, but it is a direct, multiyear commitment to inference infrastructure. Joby’s $500 million acquisition is different in character, yet its $450 million cash component makes the defense expansion a material capital-allocation decision rather than a partnership announcement.

Trend Watch

1. AI spending is spreading from GPUs into the rest of the cluster. CoreWeave’s capex and Lumentum’s optics growth point to the same systems-level reality: accelerator performance only matters if power, networking and optical links can keep the machines fed. The opportunity set is therefore widening toward interconnects, power delivery, cooling, cluster orchestration and utilization software.

2. Inference is becoming contractable infrastructure. The IBM–Together AI agreement is a useful marker because it attaches a dollar value and a deployment date to dedicated inference capacity. If this pattern persists, AI infrastructure will look less like opportunistic spot GPU renting and more like a mix of reserved capacity, managed inference and software-defined routing across multiple providers.

3. Agentic development is pushing testing infrastructure closer to the model loop. Google’s DDP preview reflects a practical problem in autonomous software work: an agent that can write code still cannot know whether a mobile experience works on a foldable screen, a particular chipset or a lower-end device without access to the environment. Real-device APIs could become a standard component of agentic CI/CD.

People & Companies to Watch

CoreWeave is the clearest company to watch for whether extreme AI infrastructure growth can produce durable operating leverage before financing costs overwhelm the model. Lumentum matters because photonics is becoming a visible constraint and spending category inside AI clusters; its next few quarters will help show whether today’s optical demand is a sustained capacity cycle or a short-lived inventory surge. Together AI is also worth monitoring: after raising $800 million in July, it is translating funding into contracted Blackwell-class capacity, which will test whether an open-model inference platform can build meaningful scale without owning all of its underlying infrastructure.

What to Watch Next

CoreWeave’s next checkpoints are capacity delivery, interest expense and the conversion rate of backlog into revenue; any slippage in power, construction or financing would weaken the bullish interpretation of demand. For Lumentum, watch whether 1.6T transceivers and high-power laser shipments continue to ramp while margins hold as supply expands. IBM and Together AI now have a concrete early-2027 deployment target, so procurement milestones, cluster scale and customer workloads will matter more than another partnership announcement.

Google’s DDP becomes more interesting once the agent skill and Android integrations are broadly usable: adoption, test reliability and per-minute economics will show whether real-device access becomes routine infrastructure for coding agents. Joby’s acquisition still has to close, and investors should look for defense program wins that demonstrate Resonant is more than a capability purchase.

The opportunity areas are similarly specific: optical networking, inference scheduling and observability, agent-oriented testing infrastructure, and dual-use autonomy systems. The main invalidation risk across the AI stories is that contracted demand grows faster than profitable utilization; across Joby, it is that defense expansion consumes capital without reducing the commercial business’s execution risk.