Tech news, 30 October 2026: the AI capex bill arrives, Samsung's HBM record, FuriosaAI, ServiceNow and an ESX zero-day

1. Google Cloud doubles, but Alphabet's cash flow turns negative

Alphabet reported Google Cloud revenue of about $31 billion for the quarter, roughly double a year earlier, while free cash flow was negative as capital spending on AI infrastructure rose¹. CNBC reported the shares were heading for a record post-earnings loss on 29 October².

CTO take: demand for cloud AI is real, and so is the bill for building it. Providers carrying that much capex will push for long commitments and premium AI services. Before signing multi-year deals, know which workloads truly need a specific provider's models or TPUs and which could move. Our note on planning around the AI capacity crunch covers how to keep that leverage.

2. AWS beats estimates as Amazon's free cash flow hits minus $42.4 billion

Amazon's third-quarter results showed AWS revenue ahead of estimates, but trailing free cash flow of negative $42.4 billion; the company held its 2026 capital spending forecast unchanged and shares fell in extended trading³.

CTO take: two of the three largest clouds are now spending more cash than they make to build AI capacity. That does not threaten service, but it does shape commercial behaviour. Expect committed-spend discounts to be tied to AI products, and make sure your AI cost allocation is good enough to tell which teams are driving the spend before renewal season.

3. Samsung posts a record quarter on AI memory

Samsung Electronics confirmed record third-quarter operating profit of about 107.4 trillion won on 29 October, driven by its chip business and high-bandwidth memory for AI servers⁴. The Next Web reports the memory unit set records for both sales and profit on server demand and higher prices⁵.

CTO take: memory, not only GPUs, is now a cost driver for AI hardware, and server memory prices have risen sharply this year. If you are buying on-premises inference kit, as in private LLMs on-premises, get quotes early and size models to the memory you can actually buy. Techniques such as speculative decoding and quantisation get more valuable when memory is the expensive part.

4. Korea backs FuriosaAI's inference chips with a $620 million round

FuriosaAI closed a $620 million Series D to scale its RNGD inference chip, which it says is in mass production, and to fund a third-generation chip; Korea's National Growth Fund is the largest investor⁶. Jon Peddie Research notes the company's performance claims come without published test conditions⁷.

CTO take: alternatives to Nvidia for inference are maturing, and governments are funding them for sovereignty reasons. Keep your serving stack portable, for example behind vLLM and an inference gateway, so you can benchmark a new accelerator on your own workloads rather than trusting vendor numbers.

5. ServiceNow raises its AI target and suffers a record share fall

ServiceNow reported third-quarter subscription revenue up 21% and raised its 2026 target for AI annual contract value to $1.7 billion⁸. The shares still fell about 19% in trading on 29 October, which Yahoo Finance links to investor fears that AI agents will erode traditional software seats⁹.

CTO take: the market is betting that agents will change how enterprise software is priced and bought. For buyers, that means more leverage on per-seat renewals and more vendors bundling agents. Ask vendors how pricing works when an agent does the work of a seat, and test their agents against your own reliability evaluations before paying for them.

6. Attackers chain an ESX zero-day and PHP2Shell to get root

Help Net Security reports that a China-nexus group used a VMware ESX logging flaw, fixed by Broadcom in July 2026, as a zero-day to hide its activity, and separately that attackers chained the PHP2Shell flaws with a Linux kernel bug called Fragnesia to gain root on servers and firewalls¹⁰.

CTO take: PHP2Shell is no longer just a web-tier problem once it is chained to a kernel flaw. If you have not finished patching after Tuesday's warning, do it now, patch kernels too, and check that hypervisor logs are shipped somewhere an attacker on the host cannot edit. Our legacy PHP estate case study shows how to triage an older estate quickly.

What I'd do this week

• Confirm ESX hosts, PHP runtimes and Linux kernels are on patched versions, and ship hypervisor logs off-host.
• List cloud commitments due for renewal in the next year and identify which workloads could move.
• Get updated quotes for any planned GPU or memory purchases; prices are moving.
• Ask SaaS vendors how agent-based pricing will work before your next renewal.

Building in Korea, the US or anywhere else? See where I work, the US page, or analyse your challenge and bring the result to a call.