AI Spending Jitters, Oracle's Chip Mix and a Firewall Exploit Worth Checking
- Madeleine Streets
- 844 views
Nasdaq 100 heads for a $1 trillion-plus drop as AI spending gets scrutiny, while PAN-OS firewall exploits put management interfaces back in focus.

Three stories from the last week of October matter for infrastructure planning. Markets started questioning what AI spending actually returns, Oracle widened its hardware options for AI workloads, and attackers chained flaws to compromise firewalls. Each one touches decisions that IT teams make in the next budget cycle.
The AI spending question reaches the stock market
Technology stocks sold off on Friday, putting the Nasdaq 100 on course for a loss of more than $1 trillion in market value. That would be its steepest potential one-day decline since mid-June. Amazon shares fell about 3% after the company reported a trailing-12-month free cash flow deficit of $42.4 billion, driven largely by AI infrastructure spending. Amazon also signalled that capital expenditure would rise again next year. Intel dropped roughly 30% after warning that weak PC demand, manufacturing problems and heavy investment in production capacity would weigh on its outlook. Alphabet shares had already sold off on Thursday, as investors weighed rising AI spending against concerns about its advertising business.
Here's the thing: the reaction isn't really about one bad quarter. It reads as impatience. Investors want proof that money going into data centres, chips and compute capacity turns into returns, and they're less willing to accept rising revenue and bigger investment forecasts at face value. My read is that this pressure will eventually reach the people buying cloud services. Providers facing demands for profitability tend to slow build-outs, change how they finance them, or revisit pricing. For CIOs, the economics behind today's cloud and AI offers could shift faster than multi-year contracts assume. Pricing and commitment terms deserve a second look now, not at renewal.
Oracle widens the chip menu
On Oct. 27, Oracle announced an expanded agreement with Nvidia to deploy Nvidia's latest Vera Rubin systems at multi-gigawatt scale. Oracle is also broadening the processors available through Oracle Cloud Infrastructure. It plans to offer Google's tensor processing units for the first time, alongside inference chips from Cerebras Systems and SambaNova.
The timing makes sense. Organisations are moving from training AI models toward running them at scale, and those two jobs have different hardware profiles. Training pushes raw throughput; inference often cares more about latency and cost per request. A cloud that offers several processor families lets customers match hardware to each workload instead of forcing everything onto one architecture.
That said, the benefit depends on execution. Availability dates, regional coverage and real-world performance against workloads you actually run will decide whether this becomes flexibility or just a longer product sheet. If your team is planning AI capacity for the coming year, ask Oracle, or any provider, for benchmark conditions and availability commitments in writing, and test your own models before committing budget. Greater hardware diversity could eventually bring more room on performance, availability and cost, but that's a "could," not a "will," until the offerings are live.
Firewall exploits show how management interfaces get chained
This is the item that needs action this week. Security researchers reported on Oct. 26 that attackers had compromised 25 servers by chaining two PHP2Shell vulnerabilities with a separate Linux kernel privilege-escalation flaw. Among the affected systems were Palo Alto Networks PAN-OS firewalls. The attackers went after internet-exposed management interfaces to gain root access and, in some cases, move deeper into internal networks. Palo Alto Networks has confirmed active exploitation of the PHP vulnerabilities and has worked with researchers at Dream Research to investigate the campaign.
The lesson is less about any single flaw and more about exposure. A management interface reachable from the internet gives an attacker a starting point, and chaining bugs can get past protections that were designed to hold each one in isolation. A firewall that sits at the edge and gets compromised is a serious problem, because it's the control point everything else depends on.
If you run PAN-OS, here's what I'd do first, in this order:
Confirm which PAN-OS versions are deployed across every firewall, including ones that were set up for a single project and forgotten. Check the vendor's guidance for the fixed releases and apply them. Restrict management interfaces to trusted administrative networks rather than leaving them reachable from the internet. Then review logs and configurations for signs of compromise, and treat anything unexpected as a reason to escalate. Don't assume that a firewall which is patched today was clean last week.
Alibaba Cloud opens in the Netherlands
Alibaba Cloud launched its first public cloud region in the Netherlands on Oct. 29, with two data centres in Amsterdam. European customers can now use Alibaba's cloud computing, storage, networking and AI services from infrastructure inside the region. The company also introduced new AI and hybrid cloud offerings as part of a wider European push. Those plans include a potential cloud region in Finland and additional data centre capacity in Germany and France.
For European businesses, this is another option in the vendor shortlist. Alibaba is positioning local infrastructure alongside its own AI models and services to compete with established hyperscalers. Whether it wins workloads will depend less on the launch itself and more on the details that matter to your organisation: service capabilities, data-residency needs, security assurances and the operational cost of running another cloud environment alongside the ones you already have. Adding a provider is never free, even when the pricing looks good.
What's next
Early November brings a useful read on demand. Palantir, Cloudflare and Atlassian report results in the Nov. 2–5 window, which should show how enterprises are spending on AI-powered software, cloud services and security products. AMD, Arista Networks, Arm and Qualcomm report on Nov. 3–4, with a focus on AI processors and the networking gear that supports data centre growth. Those numbers will tell you whether the late-October sell-off reflected a temporary mood or a shift in how investors judge AI capacity.
The common thread is that infrastructure decisions made now will be tested against tighter scrutiny of cost and returns, and against security exposure that doesn't wait for a change window. Check your firewall management exposure this week, and revisit AI capacity commitments with the assumption that pricing and availability could move.