Google readies new AI inference chips as Nvidia's grip faces a test
Google is moving beyond training into faster, cheaper inference chips—just as AI usage shifts to everyday apps. Here’s what that means for costs, vendors, and your stack.
One-Line Summary
Big platforms push deeper into AI infrastructure and agents—Google on inference chips, Adobe on marketing agents, and Siemens on factory automation—tightening the link between hardware choices and real business workflows.
Big Tech
Google eyes new chips to speed up AI results, challenging Nvidia
Google, the company behind Search, Android, and Gemini, is preparing new chips focused on AI inference—the part that answers user requests after a model is trained—to deliver results faster and at lower cost, positioning itself more directly against Nvidia in a fast-growing segment. Bloomberg reports Google builds on recent momentum after deals with Meta and Anthropic, as top AI developers increasingly procure Google’s AI chips. 1
Inference has become the cost driver for AI apps, so dedicated chips can change response speed and price-per-query for consumers and enterprises. Bloomberg Law echoes the shift, noting Google’s push to introduce inference-dedicated silicon as AI software adoption accelerates, intensifying competition with Nvidia’s dominant GPUs. For teams, this suggests more pricing and performance options when choosing cloud backends for chatbots, search, and content features. 2
Bloomberg frames Google’s chips as increasingly sought-after by peers and rivals, implying more multi-vendor architectures in AI stacks and potentially shorter lead times vs. constrained GPU supply. That could influence contract structures, SLAs, and regional capacity planning in 2026 procurement cycles. Secondary coverage underscores the competitive storyline, though details on model SKUs and timelines remain limited. 1 3 4
Separately, Google rolls out new touch-up tools in Google Photos that quickly smooth skin, remove blemishes, brighten eyes, and whiten teeth, available globally on Android devices with at least 4 GB RAM and Android 9.0 and up. This consumer feature shows how model inference shows up in everyday apps—and why inference hardware matters for latency and scale. 5
Adobe launches AI suite for corporate clients as competition heats up
Adobe, known for Photoshop and Acrobat as well as enterprise marketing software, introduces CX Enterprise—AI agents that automate and personalize digital marketing tasks such as engagement, sales, and loyalty—aiming to reassure customers and investors amid rising competition from AI-native tools. Reuters notes Adobe’s stock is down about 30% this year, and the company is partnering with Amazon, Microsoft, Anthropic, OpenAI, and Nvidia to ensure cross-platform operability. 6
The Wall Street Journal adds that CX Enterprise is positioned to help large firms use agentic AI within regulated, complex environments, and includes an agent called CX Enterprise Coworker that can coordinate other agents, gather business data, and execute a marketing plan. Adobe frames this as a critical step to compete with emerging AI-first apps, while offering governance and controls enterprises expect. 7
For marketing and CX teams, this signals tighter integration between creative assets and automated orchestration, potentially reducing manual segmentation and campaign setup time. Early buyer diligence should focus on data access, audit trails, model choice across partners, and how agent actions map to KPIs like conversion lift and CAC. 6
Industry & Biz
Thinking Machines clinches capital and major Nvidia chip supply deal
Thinking Machines, an AI startup founded by former OpenAI CTO Mira Murati, secures a multi-year Nvidia partnership that includes a significant investment and at least 1 gigawatt of next-generation processors via upcoming Vera Rubin systems, primarily for model training. The deal follows a prior about $2 billion seed round led by Andreessen Horowitz at a $12 billion valuation. 8
Reuters notes that 1 gigawatt of compute can cost around $50 billion, underlining how capital and supply-chain access define today’s AI race. Nvidia’s financing role across top startups, including recent large checks to OpenAI and Anthropic, reinforces a flywheel where funding and GPU access are closely intertwined. 8
Analysis rounds up a broader investment surge into Nvidia challengers and inference-focused chips, citing multi-hundred-million rounds across Europe and the U.S., even as Nvidia keeps advantages via R&D and acquisitions. For buyers, this means more silicon diversity on the horizon—but near-term availability and tooling still favor incumbents. 9
New Tools
Siemens Eigen Engineering Agent: AI for PLC coding and industrial workflows
Siemens releases the Eigen Engineering Agent, an AI product that performs multi-step reasoning and self-correction to autonomously execute tasks like PLC coding, HMI visualization, and device configuration inside real engineering systems, connecting directly to Siemens’ TIA Portal. Siemens says pilots show 2–5x faster workflows, up to 80% higher solution quality, and 50% greater engineering efficiency. 10
By referencing each project’s data structures, blocks, and component relationships, the agent delivers validated outputs tailored to legacy or undocumented systems, potentially reducing onboarding from weeks to days. Customers across sectors used it for SCL code creation and HMI work, with Siemens emphasizing a shift from suggestion-only AI to AI that completes work. 10
Competing moves at Hannover Messe show a wider industrial trend toward AI-orchestrated engineering: Rockwell demonstrated a prototype blending digital twins, an LLM copilot, and cloud controller design; Treon unveiled an AI-native maintenance orchestration layer to automate uptime across asset fleets. Together, these point to faster design-to-deploy cycles and reduced reliance on scarce automation engineers. 11 12
What This Means for You
As AI usage grows, inference—not training—drives your costs and user experience. Google’s push into inference chips signals more options to cut latency and price-per-query, which can expand use cases like real-time creative edits, customer support, and search within your apps or sites. Ask providers how upcoming silicon choices might change your unit economics.
Enterprise software is shifting from feature menus to agentic workflows that plan and execute tasks. Adobe’s CX Enterprise aims to coordinate marketing efforts with governance. If you handle campaigns or CRM, the practical checklists are data permissions, auditability, and how agent actions tie to measurable lift rather than vanity metrics.
On factory floors and in operations, Siemens’ agent shows how AI can finish engineering work, not just suggest it. For product and operations leaders, that implies shorter integration projects, faster ramps for new hires, and clearer ROI narratives when pitching automation upgrades to finance.
Capital and compute access remain kingmakers. Thinking Machines’ Nvidia pact underlines how tightly financing and hardware supply are linked. When vetting vendors, consider their long-term capacity plans and partnerships to avoid mid-rollout bottlenecks.
Action Items
- Map your inference spend: List top AI features by monthly queries and latency targets; ask your cloud or model vendor how inference-optimized chips could change pricing and response times.
- Trial Adobe’s CX Enterprise demo: If you run campaigns, request a demo focused on governance, data lineage, and how agent actions map to KPIs like conversion and CAC.
- Run a Siemens agent pilot brief: If you work with automation partners, share one PLC/HMI task and request a time-and-quality comparison against your current workflow.
- Stress-test vendor capacity: Ask your AI providers for 6–12 month capacity roadmaps, including chip supply partners and SLAs during peak demand.
Comments (0)