Andreas Timm

Tag: Google

  • MCP became infrastructure and Apple decided to rent cognition

    MCP became infrastructure and Apple decided to rent cognition

    What was announced

    Two announcements in the week of March 2–8, 2026 redrew the agent landscape. Anthropic’s Model Context Protocol crossed 97 million installs, with every major AI provider now shipping MCP-compatible tooling — moving the protocol from experiment to default infrastructure for tool-calling agents. Apple confirmed that the redesigned, AI-powered Siri targeted for release alongside iOS 26.4 will be powered by Google’s Gemini model running on Apple’s Private Cloud Compute. In parallel, Anthropic rolled out memory features to all Claude users and deployed Opus 4.6 as an add-in inside Microsoft PowerPoint and Excel.

    What it means

    The MCP install count makes the connectivity layer between agents and tools a solved problem at the standards level. That is a meaningful shift. For two years, the friction in shipping agents was that every tool integration was bespoke; the integration debt scaled linearly with the number of tools and the number of agents. With MCP at default-infrastructure scale, the integration cost is closer to fixed than linear, and the bottleneck moves from connectivity to orchestration and governance.

    Apple’s decision to rent cognition from Google for Siri is the more strategically loaded story. It signals that even the most vertically integrated consumer-tech company in the world has concluded that building competitive frontier-model capability inside the company is not the right capital allocation. The Private Cloud Compute envelope handles the data-sovereignty argument. The Gemini choice handles the capability argument. The combination is an explicit acknowledgment that frontier-model capability has consolidated at a tier of providers most companies will rent from, not build alongside.

    Andreas’s view

    My read on this: the agent stack is settling into a recognizable shape. Standards layer (MCP, becoming generic). Frontier-model layer (a small number of providers — OpenAI, Anthropic, Google, with regional players underneath). Application layer (where most enterprise value is created). The interesting strategic action for the next 24 months is in the application layer, where the questions are which workflows to embed, which data to expose, and which orchestration logic to own.

    I don’t think Apple’s choice is anomalous. It is the start of a wave. Companies that have been building internal frontier-model capabilities will increasingly find that the math does not work — the capex is consumer-internet scale, the talent is concentrated at three or four employers, and the capability gap to “good enough internal model” widens every six months. The economically rational answer for almost everyone is: rent the cognition, own the integration and the data envelope around it. Apple has now made that a defensible board-level position.

    The way I see it: the most important architectural question right now is whether the cognition layer (rented, frontier-model, expensive but improving exponentially) is clearly distinguished from the integration layer (owned, workflow-specific, where the moat actually lives). Where those layers are blurred, I’d expect companies to find themselves overpaying on one side and under-investing on the other. The Apple-Google deal is the clean reference architecture for how that separation can look.

    Three things I’m watching

    Three things I’m watching as this plays out:

    1. I’ll be watching whether companies architect the cognition layer and the integration layer separately — treating frontier-model providers as utilities while building proprietary infrastructure around workflow integration and the data envelope.
    2. The companies that preserve optionality will be the ones that default to MCP-compatible tooling for new agent integrations. The standards layer is no longer a strategic differentiator — the question is how quickly organizations stop treating it as one.
    3. I’ll be watching how internal frontier-model build efforts hold up against the Apple-Gemini reference case. Where differentiation rests on owning the model, I’m interested to see whether those bets come with a credible 36-month capex and capability projection — and what happens when they don’t.

    References and related signals

    • Crescendo AI: latest AI news and developments
    • Related signal: Anthropic’s Opus 4.6 PowerPoint and Excel integrations move frontier-model capability deeper into the enterprise default tooling, accelerating the rented-cognition pattern.
    • Related signal: NVIDIA GTC 2026 (March) emphasized agentic frameworks and Fortune 500 production deployments — the application layer is where the next wave of enterprise AI value is being created.
    • Related signal: 95% of generative AI pilots still fail to reach production. The connectivity layer being solved does not solve the operating-model layer.
    • Related signal: Apple choosing Gemini over OpenAI for Siri changes the competitive math for every enterprise still scoping a frontier-model partnership.
  • Hyperscaler 2026 capex hits ~$700B. Free cash flow is the variable that breaks.

    Hyperscaler 2026 capex hits ~$700B. Free cash flow is the variable that breaks.

    What was announced

    On February 6, CNBC reported that combined 2026 AI capex commitments across Amazon, Google, Microsoft, and Meta now approach $700 billion. Amazon: roughly $200 billion. Alphabet: up to $185 billion. Microsoft: increase from prior 2025 levels (analyst consensus near $99 billion FY26, ending June). Meta: budgeted $115–135 billion. Approximately 75% of the spend is AI-related — call it $450 billion of AI infrastructure in a single year, up about 36% versus 2025. Free cash flow projections for the same set of companies show meaningful compression; Amazon is forecast to turn negative, with analyst projections of negative free cash flow between $17 billion and $28 billion in 2026.

    What it means

    Capex of this magnitude rewrites the financial model for the entire frontier compute stack. The hyperscalers are no longer building toward a near-term revenue profile — they are building toward a 5-to-7-year usage curve they believe is coming. That is a different posture than the 2018–2022 capex cycle, which was largely demand-led. This one is conviction-led, and the conviction is asymmetric: if AI compute demand materializes at the projected rate, today’s capex looks conservative; if it lags by even 18 months, the depreciation schedule eats free cash flow at a rate the public markets have not yet priced.

    A second-order effect matters more for non-hyperscalers: every CIO planning AI infrastructure in 2026 is now negotiating against a supplier base whose capacity is partially already absorbed by internal hyperscaler workloads. Pricing power for capacity is structurally higher, lead times for premium GPU instances are longer, and the cost-per-token of frontier inference will move on hyperscaler margin compression rather than competition.

    Andreas’s view

    My read on this: $700 billion is not a number that resolves itself by spreadsheet logic. It resolves itself by which hyperscaler is willing to absorb the cash-flow hit longest. The strategic question inside each company is no longer “should we build” but “which competitor blinks first when the free-cash-flow line turns red on quarterly reporting.” Amazon is closest to that line. Microsoft has the strongest cash position to absorb it. Google sits in between. Meta has the most flexibility because its core ad business is funding the AI infrastructure with the lightest accounting drag.

    I don’t think the capex commitment will be revised down materially in 2026. The competitive cost of unilaterally easing off — handing GPU capacity, customer relationships, and the model-training cadence to a competitor — is too high. What will happen instead is creative financing: more debt, more partnerships with sovereign wealth and infrastructure funds, more long-term capacity contracts that move spend off the balance sheet. The capex will continue. The accounting around it will get more interesting.

    The way I see it, adjacent businesses should not assume the capacity they need will be available at the price they modeled. My expectation is that premium-tier inference and training capacity will be priced as a scarce resource for the rest of 2026 and most of 2027. Any AI roadmap that depends on flat or declining unit costs over that window has a hidden assumption built in that I think is unlikely to hold.

    Three things I’m watching

    1. I’ll be watching whether companies move to lock multi-year capacity contracts for premium inference and training now, or wait — because negotiating against scarcity in 2027 will be more expensive than over-committing modestly in 2026.
    2. The companies that preserve optionality will be the ones that have stress-tested their AI cost models against a scenario where frontier-tier compute prices are flat or rising for 18 months — and redesigned the workflow, not the budget, when the unit economics broke.
    3. Hyperscaler free-cash-flow disclosures over the next four quarters are the leading indicator I’m focused on — they will show whether the capex commitments hold or quietly compress.

    References and related signals

  • Humanoids crossed from demo to deployment in one week

    Humanoids crossed from demo to deployment in one week

    What was announced

    At CES 2026 in Las Vegas (Jan 5–9), a cluster of robotics announcements crossed the same threshold in a single week. Boston Dynamics unveiled the production-ready electric Atlas with Hyundai committing the first fleet to its Metaplant in Savannah, Georgia, and announced a partnership with Google DeepMind to integrate Gemini Robotics models into the platform. LG demonstrated CLOiD performing real household work — laundry, dishwasher loading, food preparation — in a staged living environment. EngineAI introduced the T800 with a $25,000 starting price and mid-2026 shipping. CES listed 40 companies referencing humanoids on the show floor.

    What it means

    A human factory engineer in navy work clothes works alongside a matte-white humanoid robot at a metal workbench.
    Side by side, not face to face.

    For three years humanoids were a category of demo videos. CES 2026 is where the category became a category of contracts. Production is committed, factories are named, prices are listed, and the foundation-model layer (Gemini Robotics, comparable initiatives at other labs) supplies the cognitive component that previously made every demo brittle. The constraint is no longer “can it walk on stage.” The constraint is “what does the deployment workflow look like, and who owns the integration.”

    From this follows a second-order effect: industrial buyers now have a real procurement question to answer in 2026 — not in 2030. Hyundai’s timeline (Atlas at Metaplant, dedicated robotics factory targeting 30,000 units per year by 2028) is the explicit benchmark. Every competing automaker, every large logistics operator, and every contract manufacturer now sits with a known reference deployment to react to.

    Andreas’s view

    My read on this: the news is not that the robots are good enough. The news is that buyers have decided they are good enough to commit — and the price has moved into range. At $25,000, a humanoid sits below the annual cost of an industrial worker in most developed markets. That shifts the question from “is this technology real” to “where does it amortize fastest.”

    My three takeaways:

    1. The barrier that fell was cognitive, not mechanical. The hardware has been close to ready for years. What changed is that foundation models — think Atlas plus Gemini Robotics — absorbed the cognitive deficit that kept robots out of unstructured environments. CES 2026 looks different because the system is different, not just the chassis. I think anyone framing this as “better robots” is underestimating the speed of what comes next.

    2. The 2030 humanoid timeline is already stale. In my view, this is now a 2026 pilot conversation for any organization with manufacturing, warehousing, or fulfillment in its operations footprint — anywhere unit-level labor is the dominant cost driver. Not as a capex bet, but as a learning investment. The compounding advantage goes to whoever builds operational muscle around these systems first.

    3. The real cost of waiting isn’t hardware — it’s the operating model. Hardware will be available to everyone. What won’t be available off the shelf is three years of deployment experience. My expectation is that late movers won’t just be buying machines from competitors — they’ll be importing the playbook for how to use them.

    References and related signals