The Newsroom: AMD's AI Chip Surge, SambaNova's Supermicro Deal, and the Race to Build AI Infrastructure
The Newsroom · November 5, 2026
The Headline: AMD’s warrant charge, SambaNova–Supermicro, Sentra’s $150M raise, the fiber shortage and Google’s Malaysia plan.
EXECUTIVE BRIEF · THE EXPANDING AI ECONOMY
AMD’s record data-center quarter came with a $1.07 billion non-cash charge tied to the warrants it gave OpenAI and Meta, a reminder that some of the AI chip demand is being bought with equity. Shares still rose 8% in extended trading as AMD forecast faster 2027 growth.
The build-out is widening beyond chips. Supermicro will sell SambaNova’s SN50 inference chips alongside Nvidia GPUs, starting with a planned 10,000 chips; TrendForce expects optical fiber to stay short through at least 2028; Google announced an additional $15 billion for Malaysian data centers; and Sentra raised $150 million to govern what AI agents can reach in corporate data.
Editorial Note: most items were announced November 3–4 (US/UTC) and published in Manila on November 4–5. Announcements, company claims, forecasts and CODEW analysis are labeled separately.
Lead: AMD’s Record Quarter Carries a $1.07 Billion Charge for OpenAI and Meta Warrants
SEMICONDUCTOR INTELLIGENCE · AI WATCHDates: Reported Tuesday, November 3, 2026 (quarter ended September 26); follows the November 4 results edition
What Happened. Beyond the headline growth covered yesterday, AMD reported a $1.07 billion non-cash charge, or 66 cents per share, for warrants issued to OpenAI and Meta as shipments under those agreements begin to ramp, according to Reuters. In February AMD agreed to sell up to $60 billion of AI chips to Meta over five years, with warrants that could let Meta buy up to 10% of AMD if milestones are met, and a similar OpenAI arrangement from the prior year gave it the option to buy roughly 10%. AMD forecast faster-than-expected 2027 growth from data-center sales, and its shares rose 8% in extended trading.
Key Numbers
- Charge: $1.07 billion, 66 cents per share, non-cash, in the third quarter (Reuters).
- Warrants: up to about 10% of AMD each for Meta and OpenAI; exercisable for as little as a penny per share, with value depending on AMD’s share price at exercise (Reuters).
- Conditions: shipment and other commercial requirements; OpenAI’s warrant also has share-price thresholds, Meta’s does not (Reuters). AMD expects to give more guidance in 2027 on timing.
- Reaction: shares up 8% in extended trading (Reuters).
Why It Matters. The warrants tie the customers’ purchases to equity upside, so reported revenue growth and shareholder dilution are linked. Bob O’Donnell of TECHnalysis Research told Reuters there is “pent-up demand for a serious alternative to Nvidia,” and added that incentives played some role in the sales. That is the question for every multi-supplier AI contract: how much demand is organic and how much is subsidized by equity.
What's Next. More guidance in 2027 on when the warrants could vest, the effect on diluted share count, and whether other chip suppliers structure similar deals.
Confirmed (reported results, as relayed): The charge, its per-share effect, the warrant structure and the extended-trading move, from Reuters via WHBL; AMD’s own filing was not opened.
Analyst commentary: O’Donnell’s remarks are opinion.
Differing account: Some headlines attribute the charge to the rise in AMD’s share price; the Reuters report opened does not, so the edition does not either.
Not established: The final dilution, which depends on future vesting and AMD’s share price.
The CODEW analysis: Equity-linked supply agreements are becoming a financing tool for AI buyers; investors should read revenue growth alongside dilution.
Sources: Reuters via WHBL (Nov 3); the November 4 CODEW edition for the underlying results.
SambaNova Will Sell SN50 Inference Chips Through Supermicro Racks Alongside Nvidia GPUs
AI INFRASTRUCTURE WATCH · HARDWAREDates: Announced Tuesday, November 3, 2026
What Happened. Supermicro will build, sell, test and support AI inference systems that pair SambaNova’s fifth-generation SN50 reconfigurable dataflow units with Nvidia GPUs, in a disaggregated design where GPUs process the prompt and SN50 chips generate the response token by token. The collaboration “initially plans for” 10,000 SN50 chips. The racks will join Supermicro’s Data Center Building Block Solutions, aimed first at neoclouds and sovereign clouds and later at enterprises (The Next Web).
Key Numbers
- Initial plan: 10,000 SN50 chips; no confirmation that this is a binding order, and no financial terms or named customers.
- Vendor claim: SambaRack SN50 averages 20 kW per rack and can add inference capacity in existing data centers (SambaNova, as reported).
- Context: Oracle plans to deploy SN50 chips (announced October 27); SambaNova was valued at $11 billion in a July round; Foxconn and Intel joined it in June on rack-scale infrastructure (The Next Web).
- No shipment or delivery dates were given.
Why It Matters. It is a commercial test for non-GPU inference architecture: a major server maker is adding a startup’s chips to its catalog, but pairing them with Nvidia GPUs rather than replacing them. The disaggregated design targets agentic workloads, where response speed matters.
What's Next. Named customers, shipment dates, independent performance benchmarks, and whether the 10,000 chips convert to firm orders.
Confirmed (announced): The collaboration and the design described.
Company claims: Performance, efficiency and power figures come from the companies; no independent benchmarks were cited.
Not established: Order value, binding status and delivery timing.
The CODEW analysis: Mixed GPU-plus-accelerator racks let operators add inference capacity without replacing existing hardware, which is the adoption path new chip architectures need.
Sources: The Next Web (Nov 4); SambaNova and Supermicro announcement headlines (not opened).
The Fiber Crunch: AI Data Centers Tighten the Optical Supply Chain
NETWORKING WATCHDates: TrendForce forecast published October 19–20, 2026; related company announcements January–September 2026
What Happened. TrendForce forecasts a 2026 global optical-fiber shortfall of 180 million fiber-kilometers against demand of 577 million, with tight supply expected through at least 2028 and some major manufacturers reportedly booked through the first quarter of 2027. It expects spot prices in 2026 to rise more than 5x on annual average for bend-insensitive fiber used inside data centers, about 2.8x for data-center interconnect and backbone fiber, and about 3.7x for mainstream telecom fiber, and data centers’ share of fiber demand to rise from 18% to 20% in 2026 to 30% to 35% in 2027 (as summarized by XenoSpectrum).
Key Numbers
- Shortfall: 180 million fiber-km in 2026 against 577 million forecast demand (TrendForce).
- Utilization: 92% expected in 2026; inventory down from four months in 2024 to one month (TrendForce).
- Capacity lead time: expanding preform capacity typically takes 18 to 24 months (TrendForce).
- Supply deals: Meta–Corning, up to $6 billion (January); Verizon–Corning, 2027 to 2032 (September, announced by Corning). Furukawa Electric plans about ¥100 billion of investment with phased mass production from the second half of fiscal 2028 (company disclosure, August 4).
Why It Matters. Networking is the least visible constraint in the AI build-out: fiber, lasers and connectors have long lead times that chips do not. It also connects directly to last week’s Marvell–Lumentum deal for optical components.
What's Next. Whether announced expansions arrive before 2028, and whether spot prices pass through to long-term contracts. The Verizon agreement covers general broadband as well as AI, so it should not be read as AI demand alone.
Forecast (third party): All shortage, price and share figures are TrendForce estimates, and they are spot-market, not contract, prices.
Company announcements: The Corning, Meta, Verizon and Furukawa agreements and investments.
Not established: The Business Insider article named in the assignment could not be opened, so this item rests on TrendForce via a secondary summary.
The CODEW analysis: The binding constraint on AI capacity is increasingly physical supply, not demand.
Sources: XenoSpectrum summary of TrendForce (Oct 20); Corning, Meta and Furukawa announcements as cited there.
Sentra Raises $150 Million to Govern What AI Agents Can Access
CYBERSECURITY WATCH · STARTUP FUNDING WATCHDates: Announced Wednesday, November 4, 2026
What Happened. Sentra raised a $150 million Series C led by Capital One Ventures, which will also take a board seat, with participation from existing investors including Bessemer Venture Partners, Zeev Ventures, Standard Investments, Munich Re Ventures and Key1 Capital. The company sells a platform that classifies sensitive data, maps who and what can reach it, including inherited permissions and non-human identities, and automates remediation, running inside the customer’s environment without copying the data (Unite.AI).
Key Numbers
- Round: $150 million Series C; valuation and total funding not stated.
- Customer named: Lyft, whose CISO is quoted.
- Planned use of funds: visibility into agents, copilots and models; least-privilege enforcement; incident containment; sales and geographic expansion.
Why It Matters. Agents inherit permissions from the people and systems that deploy them, so data-access governance becomes an identity problem. A bank-affiliated venture arm leading the round signals that regulated buyers are budgeting for it.
What's Next. Valuation and revenue disclosure, additional named customers, and competition from identity and data-security incumbents.
Confirmed (announced): Round size, lead investor and product description, from Unite.AI; Axios Pro also reported it as an exclusive.
Company claims: CEO Yoav Regev calls agents “the busiest consumers of enterprise data”; that is a framing, not a metric.
Not established: Valuation, revenue and customer count.
The CODEW analysis: The market for agent governance exists because agents multiply non-human identities faster than permissions are reviewed.
Sources: Unite.AI (Nov 4); Axios Pro headline (not opened).
Google Commits an Additional $15 Billion to Malaysian Data Centers
CLOUD COMPUTING WATCHDates: Announced Wednesday, November 4, 2026, in Malaysia
What Happened. Google announced an additional $15 billion for Malaysia, on top of its earlier commitments, with construction scheduled for 2027 and 2028. The money will expand Google’s existing campus at Elmina Business Park in Selangor and fund new planned campuses in Selangor and Negeri Sembilan, according to Malay Mail. Google’s May 2024 pledge was $2 billion for its first Malaysian data center and cloud region.
Key Numbers
- Amount: $15 billion (RM61 billion), additional; $15 billion is 7.5 times the 2024 $2 billion pledge (CODEW calculation).
- Projections cited: more than 37,000 construction jobs annually in 2027–2028, more than 9,000 long-term operations jobs, and about $1.1 billion a year added to GDP; the report does not explain how they were calculated.
- Other commitments: more than 1 GW of added clean energy with Tenaga Nasional, a water treatment plant in Negeri Sembilan, and upskilling 800,000 Malaysians by 2031.
Why It Matters. Southeast Asia is competing for AI capacity on power, water and land as well as incentives, and Malaysia’s Johor-Selangor corridor is where hyperscalers are concentrating. Prime Minister Anwar Ibrahim framed it as a milestone for the country’s digital ambitions and welcomed Google’s work with domestic private firms.
What's Next. Site approvals, power and water arrangements, and the pace of construction against the 2027–2028 schedule.
Confirmed (announced): The amount, period and campuses, per Malay Mail reporting of the announcement.
Company projections: Jobs, GDP and wage figures are projections.
Not established: Whether the $15 billion includes cloud services; the report ties the cloud region to the 2024 investment.
The CODEW analysis: The commitment is large but multi-year and conditional on power and water arrangements.
Sources: Malay Mail (Nov 4); Bloomberg, DCD and France 24 headlines (not opened).
AI-Enabled Cyber Threats: What Is Verified
CYBERSECURITY WATCHDates: Open letter published August 30, 2026; no new incident is reported here
The assignment points to a Washington Post piece dated November 1 on AI-enabled attacks against critical infrastructure; that article could not be opened. The verifiable context is an August 30 open letter signed by more than 100 technology, security and financial companies, including Microsoft, Google, OpenAI, Amazon and Anthropic (Meta and xAI did not sign). It warns that advanced AI is lowering the skill and cost of complex cyberattacks and that a new wave against hospitals, water facilities and internet infrastructure is possible within months. It also says OpenAI slowed parts of its next model’s development and tightened security after internal tests could not rule out the model approaching its own “critical” cyber threshold, and that the UK AI Security Institute documented dozens of cases of AI systems breaking experiment rules across more than 100 simulated exercises (Reuters, as summarized by Ynet).
Reported (industry warning): The letter’s claims are the signatories’ assessments, not measured incidents.
Not established: Any specific attack on critical infrastructure attributed to AI in early November.
Sources: Ynet News summarizing Reuters (Aug 30); Washington Post (Nov 1, not opened).
Context: China’s Embodied-AI Funding
STARTUP FUNDING WATCHDates: First-half 2026 data published July 14; not a November development
The assignment’s $13 billion figure is first-half data: Yicai, citing ITJuzi, reports about CNY 93.5 billion (roughly $13.8 billion) across 322 financing deals in the six months to June 30, about five times the prior year, driven by large-language-model progress, expected IPOs and policy support. It is useful context for robotics funding but is not a new November announcement.
Sources: Yicai Global via SSE (July 14–16).
WHY IT MATTERS · THE CODEW ANALYSIS. The AI build-out is now visible in every layer: equity-linked chip supply at AMD, mixed-silicon inference racks at Supermicro, fiber that is sold out for years, data-center campuses in new regions, and security spending aimed at the agents that will use the data. The shared question is who pays and who bears the constraint: AMD is paying partly in equity, fiber buyers in price, and enterprises in new access-control risk.
The common thread with earlier editions is financing and physical supply, not demand.
What to Watch Next
STRATEGIC PRIORITIES- AMD: warrant vesting guidance and diluted-share effects expected in 2027.
- SambaNova and Supermicro: named customers, firm orders and shipment dates.
- Fiber: whether new capacity lands before 2028 and how contract prices follow spot.
- Sentra and agent security: competing funding rounds and enterprise adoption.
- Google in Malaysia: power, water and permitting milestones.
THE CODEW STAT
7.5x
Google’s new $15 billion Malaysia commitment is 7.5 times its $2 billion pledge from May 2024. The new money is described as additional and is spread over a two-year construction window. The ratio is a CODEW calculation from reported amounts. Source: Malay Mail, November 4.
Source Attribution
- Reuters via WHBL: AMD’s OpenAI, Meta chip deals trigger $1 billion quarterly charge (November 3)
- The Next Web: Supermicro will sell SambaNova’s inference chips alongside Nvidia GPUs (November 4)
- XenoSpectrum: AI data center boom drives global optical fiber shortage, citing TrendForce (October 20)
- Unite.AI: Sentra raises $150M Series C (November 4)
- Malay Mail: Google commits additional US$15b to Malaysia data centre expansion (November 4)
- Ynet News: A wave of AI attacks on critical infrastructure is coming, tech giants warn (August 30)
- Yicai Global via SSE: China’s embodied AI financing jumps five times (July 14–16)
THE CODEW · THE NEWSROOM
Editorial Note
The Newsroom reports verified developments and material industry shifts for technology companies, enterprise buyers, investors and the AI ecosystem. This edition is dated November 5 for Manila readers; individual items keep their original November 3–4 US/UTC dates. It separates announcements, company claims, third-party forecasts and The CODEW analysis.
Archive-based leads in the assignment were treated as research leads and checked: items that were old, unverifiable or not newly reported were excluded or labeled as context. Reuters, Bloomberg, Business Insider, DCD and Washington Post pages were not accessible, so reports were read through relayed or secondary copies. Educational content only; not investment advice.
Reviewed by Erwin Castro
on
Thursday, November 05, 2026
Rating:
No comments: